Posts in category: Google AI

Posted by Huiyu Wang, Student Researcher, and Yukun Zhu, Software Engineer, Google Research

The success of convolutional neural networks (CNNs) mainly comes from two properties of convolution: translation equivariance and locality. Translation equivariance, although not exact, ensures that the model functions well for objects at different positions in an image or for images of different sizes. Locality ensures efficient computation, but at the cost of making the modeling of long-range spatial relations challenging for panoptic segmentation of large images. For example, segmenting a large object requires modeling the shape of it, which could potentially cover a very large pixel area, and context that could be helpful for segmenting the object may come from farther away. In such cases, the inability to inform the model from context far from the convolution kernel could negatively impact the performance.

A rich set of literature has discussed approaches to solving the limitation of locality and enabling long-range interactions in CNNs. Some employ atrous convolutions, or image pyramids, which expand the receptive field somewhat, but it is still limited to a small local region. Another line of work adopts self-attention mechanisms, e.g., non-local neural networks, which allow the receptive field to cover the entire input image, as opposed to local convolutions. Unfortunately, such approaches are computationally expensive, especially for large inputs. Recent works enable building fully attentional models, but at a cost of applying local constraints to non-local neural networks. These restrictions limit the model receptive field, which is harmful to tasks such as segmentation, especially on high-resolution inputs.

In our recent ECCV 2020 paper, “Axial-DeepLab: Stand-Alone Axial-Attention for Panoptic Segmentation”, we propose to adopt axial-attention (or criss-cross attention), which recovers large receptive field in fully attentional models. The core idea is to separate 2D attention into two steps that apply 1D attention in the height and width axes sequentially. The efficiency of this approach enables attention over large regions, allowing models that learn long-range, or even global, interactions. Additionally, we propose a novel formulation for self-attention modules, which is more sensitive to the position of relevant context in a large receptive field with marginal costs. We evaluate our position-sensitive axial-attention method on panoptic segmentation by applying it to Panoptic-DeepLab, a simple and efficient method for panoptic segmentation. The effectiveness of our model is demonstrated on ImageNet, COCO, and Cityscapes. Axial-DeepLab achieves state-of-the-art results on panoptic segmentation and semantic segmentation, outperforming Panoptic-DeepLab by a large margin.

Axial-Attention Architecture
Axial-DeepLab consists of an Axial-ResNet backbone and Panoptic-DeepLab output heads, which produce panoptic segmentation results. Our Axial-ResNet is built on a ResNet architecture, in which all the 3×3 local convolutions in the ResNet bottleneck blocks are replaced by our proposed global position-sensitive axial-attention, thus enabling both a large receptive field and precise positional information.

An axial-attention block consists of two position-sensitive axial-attention layers operating along height- and width-axis sequentially.

The Axial-DeepLab height axial attention layer provides 1-dimensional self-attention globally, propagating information within individual columns — it does not transfer information between columns. The second 1D attention layer operating in the horizontal direction allows one to capture both column-wise and row-wise information. This separation reduces the complexity of self-attention from quadratic (2D) to linear (1D), which enables using a much larger (65×65 vs. previously 3×3) or even global context in all layers for long-range modeling in panoptic segmentation.

A message can be passed globally with two hops.

Note that a message or feature vector at (x1, y1) can always be passed globally on a 2D lattice to any position (x2, y2), with one hop on the height-axis (x1, y1 →x1, y2), followed by another hop on the width axis (x1, y2 → x2, y2). In this way, we are able to model 2D long-range relations in a single residual block. This axial-attention design also reduces the complexity from quadratic to linear and enables global receptive fields in all layers of a model.

Position-Sensitive Self-Attention
Additionally, we propose a position-sensitive formulation for self-attention. Previous self-attention formulations enabled a given pixel A to aggregate long-range context B, but provided no information about where in the receptive field the context originated. For example, perhaps the feature at pixel A represents the eye of a cat, and the context B might be the nose and another eye. In this case, the aggregated feature at pixel A would be a nose and two eyes, regardless of the geometric structure of a face. This could cause a false indication of the presence of a face when the two eyes are on the bottom-left of an image and the nose is on the top-right. A recently proposed solution is to impose a positional bias on where in the receptive field the context can originate. This bias depends on the feature at A only, (an eye), but not the feature at B, which contains important contextual information.

In this work, we let this bias also depend on the context feature at B (i.e., the nose and another eye). This change enables a more accurate positional bias when a pixel and the context informing it are far away from one another and thus contains different information about the bias. In addition, when pixel A aggregates the context feature B, we also include a feature that indicates the relative position from A to B. This change enables A to know precisely where B originated. These two changes make self-attention position-sensitive, especially in the situation of long-range modeling.

Results
We have tested Axial-DeepLab on COCO, and Cityscapes for panoptic segmentation. Improvements over the state-of-the-art Panoptic-DeepLab for each dataset can be seen in the table below. In particular, our Axial-DeepLab outperforms Panoptic-DeepLab by 2.8% Panoptic Quality (PQ) on the COCO test-dev set. Our single-scale small model performs better than multi-scale Panoptic-DeepLab while improving computational efficiency by 27x and using only 1/4 the number of parameters. We also show state-of-the-art results on Cityscapes. Moreover, we find that the performance increases as the block receptive field increases from 5 × 5 to 65 × 65. Our model is also more robust to out-of-distribution scales, on which the model was not trained.

Besides our main results on panoptic segmentation, our full axial-attention model, Axial-ResNet, also performs better than the previous best stand-alone self-attention model on ImageNet.

Conclusion
We have proposed and demonstrated the effectiveness of position-sensitive axial-attention on image classification and panoptic segmentation. On ImageNet, our Axial-ResNet, formed by stacking axial-attention blocks, achieves state-of-the-art results among stand-alone self-attention models. We further convert Axial-ResNet to Axial-DeepLab for bottom-up panoptic segmentation, and also show state-of-the-art performance on several benchmarks, including COCO, and Cityscapes. We hope our promising results could establish that axial-attention is an effective building block for modern computer vision models.

Acknowledgements
This post reflects the work of the authors as well as Bradley Green, Hartwig Adam, Alan Yuille, and Liang-Chieh Chen. We also thank Niki Parmar for discussion and support; Ashish Vaswani, Xuhui Jia, Raviteja Vemulapalli, Zhuoran Shen for their insightful comments and suggestions; Maxwell Collins and Blake Hechtman for technical support.

Read More

Posted by Natasha Noy, Research Scientist; and Omar Benjelloun, Software Engineer, Google Research

There are tens of millions of datasets on the web, with content ranging from sensor data and government records, to results of scientific experiments and business reports. Indeed, there are datasets for almost anything one can imagine, be it diets of emperor penguins or where remote workers live. More than two years ago, we undertook an effort to design a search engine that would provide a single entry point to these millions of datasets and thousands of repositories. The result is Dataset Search, which we launched in beta in 2018 and fully launched in January 2020. In addition to facilitating access to data, Dataset Search reconciles and indexes datasets using the metadata descriptions that come directly from the dataset web pages using schema.org structure.

As of today, the complete Dataset Search corpus contains more than 31 million datasets from more than 4,600 internet domains. About half of these datasets come from .com domains, but .org and governmental domains are also well represented. The graph below shows the growth of the corpus over the last two years, and while we still don’t know what fraction of datasets on the web are currently in Dataset Search, the number continues to grow steadily.

Growth in the number of datasets indexed by Dataset Search

To better understand the breadth and utility of the datasets made available through Dataset Search, we published “Google Dataset Search by the Numbers”, accepted at the 2020 International Semantic Web Conference. Here we provide an overview of the available datasets, present metrics and insights originating from their analysis, and suggest best practices for publishing future scientific datasets. In order to enable other researchers to build analysis and tools using the metadata, we are also making a subset of the data publicly available.

A Range of Dataset Topics
In order to determine the distribution of topics covered by the datasets, we infer the research category based on dataset titles and descriptions, as well as other text on the dataset Web pages. The two most common topics are geosciences and social sciences, which account for roughly 45% of the datasets. Biology is a close third at ~15%, followed by a roughly even distribution for other topics, including computer science, agriculture, and chemistry, among others.

Distribution of dataset topics

In our initial efforts to launch Dataset Search, we reached out to specific communities, which was key to bootstrapping widespread use of the corpus. Initially, we focused on geosciences and social sciences, but since then, we have allowed the corpus to grow organically. We were surprised to see that the fields associated with the communities we reached out to early on are still dominating the corpus. While their early involvement certainly contributes to their prevalence, there may be other factors involved, such as differences in culture across communities. For instance, geosciences have been particularly successful in making their data findable, accessible, interoperable, and reusable (FAIR), a core component to reducing barriers for access.

Making Data Easily Citable and Reusable
There is a growing consensus among researchers across scientific disciplines that it is important to make datasets available, to publish details relevant to their use, and to cite them when they are used. Many funding agencies and academic publishers require proper publication and citation of data.

Peer-reviewed journals such as Nature Scientific Data are dedicated to publishing valuable datasets, and efforts such as DataCite provide digital object identifiers (DOIs) for them. Resolution services (e.g., identifiers.org) also provide persistent, de-referenceable identifiers, allowing for easy citation, which is key to making datasets widely available in scientific discourse. Unfortunately, we found that only about 11% of the datasets in the corpus (or ~3M) have DOIs. We chose this subset from the dataset corpus to be included in our open-source release. From this collection, about 2.3M datasets come from two sites, datacite.org and figshare.com:

Publishers can specify access requirements for a dataset via schema.org metadata properties, including details of the license and information indicating whether or not the dataset is accessible for free. Only 34% of datasets specify license information, but when no license is specified, users cannot make any assumptions on whether or not they are allowed to reuse the data. Thus, adding licensing information, and, ideally, adding as open a license as possible, will greatly improve the reusability of the data.

Among the datasets that did specify a license, we were able to recognize a known license in 72% of cases. Those licenses include Open Government licenses for the UK and Canada, Creative Commons licenses, and several Public Domain licenses (e.g., Public Domain Mark 1.0). We found 89.5% of these datasets to either be accessible for free or use a license that allows redistribution, or both. And of these open datasets, 5.6M (91%) allow commercial reuse.

Another critical component of data reusability is providing downloadable data, yet only 44% of datasets specify download information in their metadata. A possible explanation for this surprisingly low value is that webmasters (or dataset-hosting platforms) fear that exposing the data download link through schema.org metadata may lead search engines or other applications to give their users direct access to download the data, thus “stealing” traffic from their website. Another concern may be that data needs the proper context to be used appropriately (e.g., methodology, footnotes, and license information), and providers feel that only their web pages can give the complete picture. In Dataset Search, we do not show download links as part of dataset metadata so that users must go to the publisher’s website to download the data, where they will see the full context for the dataset.

What Do Users Access?
Finally, we examine how Dataset Search is being used. Overall, 2.1M unique datasets from 2.6K domains appeared in the top 100 Dataset Search results over 14 days in May 2020. We find that the distribution of topics being queried is different from that of the corpus as a whole. For instance, geoscience takes up a much smaller fraction, and conversely, biology and medicine represent a larger fraction relative to their share of the corpus. This result is likely explained by the timing of our analysis, as it was performed during the first weeks of the COVID-19 pandemic.

Distribution of topics covered by datasets that appear in search results

Best Practices for Publishing Scientific Datasets
Based on our analysis, we have identified a set of best practices that can improve how datasets are discovered, reused and cited.

• Discoverability
  Dataset metadata should be on pages that are accessible to web crawlers and that provide metadata in machine-readable formats in order to improve discoverability.

• Persistence
  Publishing metadata on sites that are likely to be more persistent than personal web pages will facilitate data reuse and citation. Indeed, during our analysis of Dataset Search, we noted a very high rate of turnover — many URLs that hosted a dataset one day did not have it a few weeks or months later. Data repositories, such as Figshare, Zenodo, DataDryad, Kaggle Datasets and many others, are a good way to ensure dataset persistence. Many of these repositories have agreements with libraries to preserve data in perpetuity.

• Provenance
  With datasets often published in multiple repositories, it would be useful for repositories to describe the provenance information more explicitly in the metadata. The provenance information helps users understand who collected the data, where the primary source of the dataset is, or how it might have changed.

• Licensing
  Datasets should include licensing information, ideally in a machine-readable format. Our analysis indicates that when dataset providers select a license, they tend to choose a fairly open one. So, encouraging and enabling scientists to choose licenses for their data will result in many more datasets being openly available.

• Assigning persistent identifiers (such as DOIs)
  DOIs are critical for long-term tracking and useability. Not only do these identifiers allow for much easier citation of datasets and version tracking, they are also dereferenceable: if a dataset moves, the identifier can point to a different location.

Releasing Metadata for Datasets with Persistent Identifiers
As part of the announcement today, we are also releasing a subset of our corpus for others to use. It contains the metadata for more than three million datasets that have DOIs and other types of persistent identifiers –- these are the datasets that are the most easily citable. Researchers can use this metadata to perform deeper analysis or to build their own applications using this data. For example, much of the growth of DOI usage appears to have been within the last decade. How does this timeframe relate to the datasets covered in the corpus? Is the DOI usage distribution uniform across datasets, or are there significant differences between research communities?

We will update the dataset on a regular basis. Finally, we hope that focusing this data release on datasets with persistent citable identifiers will encourage more data providers to describe their datasets in more detail and to make them more easily citable.

In conclusion, we hope that having data more discoverable through tools such as Google’s Dataset Search will encourage scientists to share their data more broadly and do it in a way that makes data truly FAIR.

Acknowledgments
This post reflects the work of the entire Dataset Search team. We are grateful to Shiyu Chen, Dimitris Paparas, Katrina Sostek, Yale Cong, Marc Najork, and Chris Gorgolewski for their contributions. We would also like to thank Hal Varian for suggesting this analysis and for many helpful ideas.

Read More

This week, the 16th European Conference on Computer Vision (ECCV2020) begins, a premier forum for the dissemination of research in computer vision and related fields. Being held virtually for the first time this year, Google is proud to be an ECCV2020 Platinum Partner and is excited to share our research with the community with nearly 50 accepted publications, alongside several tutorials and workshops.

If you are registered for ECCV this year, please visit our virtual booth in the Platinum Exhibition Hall to learn more about the research we’re presenting at ECCV 2020, including some demos and opportunities to connect with our researchers. You can also learn more about our contributions below (Google affiliations in bold).

Organizing Committee
General Chairs: Vittorio Ferrari, Bob Fisher, Cordelia Schmid, Emanuele TrucoAcademic Demonstrations Chair: Thomas Mensink

Accepted Publications
NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis (Honorable Mention Award)
Ben Mildenhall, Pratul Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, Ren Ng

Quaternion Equivariant Capsule Networks for 3D Point Clouds
Yongheng Zhao, Tolga Birdal, Jan Eric Lenssen, Emanuele Menegatti, Leonidas Guibas, Federico Tombari

SoftpoolNet: Shape Descriptor for Point Cloud Completion and Classification
Yida Wang, David Joseph Tan, Nassir Navab, Federico Tombari

Combining Implicit Function Learning and Parametric Models for 3D Human Reconstruction
Bharat Lal Bhatnagar, Cristian Sminchisescu, Christian Theobalt, Gerard Pons-Moll

CoReNet: Coherent 3D scene reconstruction from a single RGB image
Stefan Popov, Pablo Bauszat, Vittorio Ferrari

Adversarial Generative Grammars for Human Activity Prediction
AJ Piergiovanni, Anelia Angelova, Alexander Toshev, Michael S. Ryoo

Self6D: Self-Supervised Monocular 6D Object Pose Estimation
Gu Wang, Fabian Manhardt, Jianzhun Shao, Xiangyang Ji, Nassir Navab, Federico Tombari

Du2Net: Learning Depth Estimation from Dual-Cameras and Dual-Pixels
Yinda Zhang, Neal Wadhwa, Sergio Orts-Escolano, Christian Häne, Sean Fanello, Rahul Garg

What Matters in Unsupervised Optical Flow
Rico Jonschkowski, Austin Stone, Jonathan T. Barron, Ariel Gordon, Kurt Konolige, Anelia Angelova

Appearance Consensus Driven Self-Supervised Human Mesh Recovery
Jogendra N. Kundu, Mugalodi Rakesh, Varun Jampani, Rahul M. Venkatesh, R. Venkatesh Babu

Fashionpedia: Ontology, Segmentation, and an Attribute Localization Dataset
Menglin Jia, Mengyun Shi, Mikhail Sirotenko, Yin Cui, Claire Cardie, Bharath Hariharan, Hartwig Adam, Serge Belongie

PointMixup: Augmentation for Point Clouds
Yunlu Chen, Vincent Tao Hu, Efstratios Gavves, Thomas Mensink, Pascal Mettes1, Pengwan Yang, Cees Snoek

Connecting Vision and Language with Localized Narratives (see our blog post)
Jordi Pont-Tuset, Jasper Uijlings, Soravit Changpinyo, Radu Soricut, Vittorio Ferrari

Big Transfer (BiT): General Visual Representation Learning (see our blog post)
Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai, Joan Puigcerver, Jessica Yung, Sylvain Gelly, Neil Houlsby

View-Invariant Probabilistic Embedding for Human Pose
Jennifer J. Sun, Jiaping Zhao, Liang-Chieh Chen, Florian Schroff, Hartwig Adam, Ting Liu

Axial-DeepLab: Stand-Alone Axial-Attention for Panoptic Segmentation
Huiyu Wang, Yukun Zhu, Bradley Green, Hartwig Adam, Alan Yuille, Liang-Chieh Chen

Mask2CAD: 3D Shape Prediction by Learning to Segment and Retrieve
Weicheng Kuo, Anelia Angelova, Tsung-Yi Lin, Angela Dai

A Generalization of Otsu’s Method and Minimum Error Thresholding
Jonathan T. Barron

Learning to Factorize and Relight a City
Andrew Liu, Shiry Ginosar, Tinghui Zhou, Alexei A. Efros, Noah Snavely

Weakly Supervised 3D Human Pose and Shape Reconstruction with Normalizing Flows
Andrei Zanfir, Eduard Gabriel Bazavan, Hongyi Xu, Bill Freeman, Rahul Sukthankar, Cristian Sminchisescu

Multi-modal Transformer for Video Retrieval
Valentin Gabeur, Chen Sun, Karteek Alahari, Cordelia Schmid

Generative Latent Textured Proxies for Category-Level Object Modeling
Ricardo Martin Brualla, Sofien Bouaziz, Matthew Brown, Rohit Pandey, Dan B Goldman

Neural Design Network: Graphic Layout Generation with Constraints
Hsin-Ying Lee*, Lu Jiang, Irfan Essa, Phuong B Le, Haifeng Gong, Ming-Hsuan Yang, Weilong Yang

Neural Articulated Shape Approximation
Boyang Deng, Gerard Pons-Moll, Timothy Jeruzalski, JP Lewis, Geoffrey Hinton, Mohammad Norouzi, Andrea Tagliasacchi

Uncertainty-Aware Weakly Supervised Action Detection from Untrimmed Videos
Anurag Arnab, Arsha Nagrani, Chen Sun, Cordelia Schmid

Beyond Controlled Environments: 3D Camera Re-Localization in Changing Indoor Scenes
Johanna Wald, Torsten Sattler, Stuart Golodetz, Tommaso Cavallari, Federico Tombari

Consistency Guided Scene Flow Estimation
Yuhua Chen, Luc Van Gool, Cordelia Schmid, Cristian Sminchisescu

Continuous Adaptation for Interactive Object Segmentation by Learning from Corrections
Theodora Kontogianni*, Michael Gygli, Jasper Uijlings, Vittorio Ferrari

SimPose: Effectively Learning DensePose and Surface Normal of People from Simulated Data
Tyler Lixuan Zhu, Per Karlsson, Christoph Bregler

Learning Data Augmentation Strategies for Object Detection
Barret Zoph, Ekin Dogus Cubuk, Golnaz Ghiasi, Tsung-Yi Lin, Jonathon Shlens, Quoc V Le

Streaming Object Detection for 3-D Point Clouds
Wei Han, Zhengdong Zhang, Benjamin Caine, Brandon Yang, Christoph Sprunk, Ouais Alsharif, Jiquan Ngiam, Vijay Vasudevan, Jonathon Shlens, Zhifeng Chen

Improving 3D Object Detection through Progressive Population Based Augmentation
Shuyang Cheng, Zhaoqi Leng, Ekin Dogus Cubuk, Barret Zoph, Chunyan Bai, Jiquan Ngiam, Yang Song, Benjamin Caine, Vijay Vasudevan, Congcong Li, Quoc V. Le, Jonathon Shlens, Dragomir Anguelov

An LSTM Approach to Temporal 3D Object Detection in LiDAR Point Clouds
Rui Huang, Wanyue Zhang, Abhijit Kundu, Caroline Pantofaru, David A Ross, Thomas Funkhouser, Alireza Fathi

BigNAS: Scaling Up Neural Architecture Search with Big Single-Stage Models
Jiahui Yu, Pengchong Jin, Hanxiao Liu, Gabriel Bender, Pieter-Jan Kindermans, Mingxing Tan, Thomas Huang, Xiaodan Song, Ruoming Pang, Quoc Le

Memory-Efficient Incremental Learning Through Feature Adaptation
Ahmet Iscen, Jeffrey Zhang, Svetlana Lazebnik, Cordelia Schmid

Virtual Multi-view Fusion for 3D Semantic Segmentation
Abhijit Kundu, Xiaoqi Yin, Alireza Fathi, David A Ross, Brian E Brewington, Thomas Funkhouser, Caroline Pantofaru

Efficient Scale-permuted Backbone with Learned Resource Distribution
Xianzhi Du, Tsung-Yi Lin, Pengchong Jin, Yin Cui, Mingxing Tan, Quoc V Le, Xiaodan Song

RetrieveGAN: Image Synthesis via Differentiable Patch Retrieval
Hung-Yu Tseng*, Hsin-Ying Lee*, Lu Jiang, Ming-Hsuan Yang, Weilong Yang

Graph convolutional networks for learning with few clean and many noisy labels
Ahmet Iscen, Giorgos Tolias, Yannis Avrithis, Ondrej Chum, Cordelia Schmid

Deep Positional and Relational Feature Learning for Rotation-Invariant Point Cloud Analysis
Ruixuan Yu, Xin Wei, Federico Tombari, Jian Sun

Federated Visual Classification with Real-World Data Distribution
Tzu-Ming Harry Hsu, Hang Qi, Matthew Brown

Joint Bilateral Learning for Real-time Universal Photorealistic Style Transfer
Xide Xia, Meng Zhang, Tianfan Xue, Zheng Sun, Hui Fang, Brian Kulis, Jiawen Chen

AssembleNet++: Assembling Modality Representations via Attention Connections
Michael S. Ryoo, AJ Piergiovanni, Juhana Kangaspunta, Anelia Angelova

Naive-Student: Leveraging Semi-Supervised Learning in Video Sequences for Urban Scene Segmentation
Liang-Chieh Chen, Raphael Gontijo-Lopes, Bowen Cheng, Maxwell D. Collins, Ekin D. Cubuk, Barret Zoph, Hartwig Adam, Jonathon Shlens

AttentionNAS: Spatiotemporal Attention Cell Search for Video Classification
Xiaofang Wang, Xuehan Xiong, Maxim Neumann, AJ Piergiovanni, Michael S. Ryoo, Anelia Angelova, Kris M. Kitani, Wei Hua

Unifying Deep Local and Global Features for Image Search
Bingyi Cao, Andre Araujo, Jack Sim

Pillar-based Object Detection for Autonomous Driving
Yue Wang, Alireza Fathi, Abhijit Kundu, David Ross, Caroline Pantofaru, Tom Funkhouser, Justin Solomon

Improving Object Detection with Selective Self-supervised Self-training
Yandong Li, Di Huang, Danfeng Qin, Liqiang Wang, Boqing Gong

Environment-agnostic Multitask Learning for Natural Language Grounded NavigationXin Eric Wang*, Vihan Jain, Eugene Ie, William Yang Wang, Zornitsa Kozareva, Sujith Ravi

SimAug: Learning Robust Representations from Simulation for Trajectory Prediction
Junwei Liang, Lu Jiang, Alex Hauptmann

Tutorials
New Frontiers for Learning with Limited Labels or Data
Organizers: Shalini De Mello, Sifei Liu, Zhiding Yu, Pavlo Molchanov, Varun Jampani, Arash Vahdat, Animashree Anandkumar, Jan Kautz

Weakly Supervised Learning in Computer Vision
Organizers: Seong Joon Oh, Rodrigo Benenson, Hakan Bilen

Workshops
Joint COCO and LVIS Recognition Challenge
Organizers: Alexander Kirillov, Tsung-Yi Lin, Yin Cui, Matteo Ruggero Ronchi, Agrim Gupta, Ross Girshick, Piotr Dollar

4D Vision
Organizers: Anelia Angelova, Vincent Casser, Jürgen Sturm, Noah Snavely, Rahul Sukthankar

GigaVision: When Gigapixel Videography Meets Computer Vision
Organizers: Lu Fang, Shengjin Wang, David J. Brady, Feng Yang

Advances in Image Manipulation Workshop and Challenges
Organizers: Radu Timofte, Andrey Ignatov, Luc Van Gool, Wangmeng Zuo, Ming-Hsuan Yang, Kyoung Mu Lee, Liang Lin, Eli Shechtman, Kai Zhang, Dario Fuoli, Zhiwu Huang, Martin Danelljan, Shuhang Gu, Ming-Yu Liu, Seungjun Nah, Sanghyun Son, Jaerin Lee, Andres Romero, ETH Zurich, Hannan Lu, Ruofan Zhou, Majed El Helou, Sabine Süsstrunk, Roey Mechrez, BeyondMinds & Technion, Pengxu Wei, Evangelos Ntavelis, Siavash Bigdeli

Robust Vision Challenge 2020
Organizers:Oliver Zendel, Hassan Abu Alhaija, Rodrigo Benenson, Marius Cordts, Angela Dai, Xavier Puig Fernandez, Andreas Geiger, Niklas Hanselmann, Nicolas Jourdan, Vladlen Koltun, Peter Kontschider, Alina Kuznetsova, Yubin Kang, Tsung-Yi Lin, Claudio Michaelis, Gerhard Neuhold, Matthias Niessner, Marc Pollefeys, Rene Ranftl, Carsten Rother, Torsten Sattler, Daniel Scharstein, Hendrik Schilling, Nick Schneider, Jonas Uhrig, Xiu-Shen Wei, Jonas Wulff, Bolei Zhou

“Deep Internal Learning”: Training with no prior examples
Organizers: Michal Irani,Tomer Michaeli, Tali Dekel, Assaf Shocher, Tamar Rott Shaham

Instance-Level Recognition
Organizers: Andre Araujo, Cam Askew, Bingyi Cao, Ondrej Chum, Bohyung Han, Torsten Sattler, Jack Sim, Giorgos Tolias, Tobias Weyand, Xu Zhang

Women in Computer Vision Workshop (WiCV) (Platinum Sponsor)
Panel Participation: Dina Damen, Sanja Fiddler, Zeynep Akata, Grady Booch, Rahul Sukthankar

*Work performed while at Google

Read More

Posted by Yonglong Tian, Student Researcher and Chen Sun, Staff Research Scientist, Google Research

Most people take for granted the ability to view an object from several different angles, but still recognize that it’s the same object— a dog viewed from the front is still a dog when viewed from the side. While people do this naturally, computer scientists need to explicitly enable machines to learn representations that are view-invariant, with the goal of seeking robust data representations that retain information that is useful to downstream tasks.

Of course, in order to learn these representations, manually annotated training data can be used. However, as in many cases such annotations aren’t available, which gives rise to a series of self- and crossmodal supervised approaches that do not require manually annotated training data. Currently, a popular paradigm for training with such data is contrastive multiview learning, where two views of the same scene (for example, different image channels, augmentations of the same image, and video and text pairs) will tend to converge in representation space while two views of different scenes diverge. Despite their success, one important question remains: “If one doesn’t have annotated labels readily available, how does one select the views to which the representations should be invariant?” In other words, how does one identify an object using information that resides in the pixels of the image itself, while still remaining accurate when that image is viewed from disparate viewpoints?

In “What makes for good views for contrastive learning”, we use theoretical and empirical analysis to better understand the importance of view selection, and argue that one should reduce the mutual information between views while keeping task-relevant information intact. To verify this hypothesis, we devise unsupervised and semi-supervised frameworks that learn effective views by aiming to reduce their mutual information. We also consider data augmentation as a way to reduce mutual information, and show that increasing data augmentation indeed leads to decreasing mutual information while improving downstream classification accuracy. To encourage further research in this space, we have open-sourced the code and pre-trained models.

The InfoMin Hypothesis
The goal of contrastive multiview learning is to learn a parametric encoder, whose output representations can be used to discriminate between pairs of views with the same identities, and pairs with different identities. The amount and type of information shared between the views determines how well the resulting model performs on downstream tasks. We hypothesize that the views that yield the best results should discard as much information in the input as possible except for the task relevant information (e.g., object labels), which we call the InfoMin principle.

Consider the example below in which two patches of the same image represent the different “views”. The training objective is to identify that the two views belong to the same image. It is undesirable to have views that share too much information, for example, where low-level color and texture cues can be exploited as “shortcuts” (left), or to have views that share too little information to identify that they belong to the same image (right). Rather, views at the “sweet spot” share the information related to downstream tasks, such as patches corresponding to different parts of the panda for an object classification task (center).

An illustration of three regimes of information captured during contrastive multiview learning. Views should not share too much information (left) or too little information (right), but should find an optimal mix (the “sweet spot”, middle) that maximizes the downstream performance.

A Unified View on Contrastive Learning
We design several sets of experiments to verify the InfoMin hypothesis, motivated by the fact that there are simple ways to control the mutual information shared between views without any supervision. For example, we can sample different patches from the same images, and reduce their mutual information simply by increasing the distance between the patches. Here, we estimate the mutual information using InfoNCE (I_NCE), which is a quantitative measure of the mutual information lower bound._. Indeed, we observe a reverse U-shape curve: as mutual information is reduced, the downstream task accuracy first increases and then begins to decrease.

Downstream classification accuracy on STL-10 (left) and CIFAR-10 (right) by applying linear classifiers on representations learned with contrastive learning. Same as the previous illustration, the views are sampled as different patches from the same images. Increasing the Euclidean distance between patches leads to decreasing mutual information. A reverse U-shape curve between classification accuracy and INCE (patch distance) is observed.

Furthermore, we demonstrate that several state-of-the-art contrastive learning methods (InstDis, MoCo, CMC, PIRL, SimCLR and CPC) can be unified through the perspective of view selection: despite the differences in architecture, objective and engineering details, all recent contrastive learning methods create two views that implicitly follow the InfoMin hypothesis, where the information shared between views are controlled by the strength of data augmentation. Motivated by this, we propose a new set of data augmentations, which outperforms the prior state of the art, SimCLR, by nearly 4% on the ImageNet linear readout benchmark. We also found that transferring our unsupervised pre-trained models to object detection and instance segmentation consistently outperforms ImageNet pre-training.

Learning to Generate Views
In our work, we design unsupervised and semi-supervised methods that synthesize novel views following the InfoMin hypothesis. We learn flow-based models that transfer natural color spaces into novel color spaces, from which we split the channels to get views. For the unsupervised setup, the view generators are optimized to minimize the InfoNCE bound between views. As shown in the results below, we observe a similar reverse U-shape trend while minimizing the InfoNCE bound.

View generators learned by unsupervised (left) and semi-supervised (right) objectives.

To reach the sweet spot without overly minimizing mutual information, we can use the semi-supervised setup and guide the view generator to retain label information. As expected, all learned views are now centered around the sweet spot, no matter what the input color space is.

Code and Pretrained Models
To accelerate research in self-supervised contastive learning, we are excited to share the code and pretrained models of InfoMin with the academic community. They can be found here.

Acknowledgements
The core team includes Yonglong Tian, Chen Sun, Ben Poole, Dilip Krishnan, Cordelia Schmid and Phillip Isola. We would like to thank Kevin Murphy for insightful discussion; Lucas Beyer for feedback on the manuscript; and the Google Cloud team for computation support.

Read More

Posted by George Tucker and Sergey Levine, Research Scientists, Google Research

Over the past several years, there has been a surge of interest in reinforcement learning (RL) driven by its high-profile successes in game playing and robotic control. However, unlike supervised learning methods, which learn from massive datasets that are collected once and then reused, RL algorithms use a trial-and-error feedback loop that requires active interaction during learning, collecting data every time a new policy is learned. This approach is prohibitive in many real-world settings, such as healthcare, autonomous driving, and dialogue systems, where trial-and-error data collection can be costly, time consuming, or even irresponsible. Even for problems where some active data collection can be used, the requirement for interactive collection limits dataset size and diversity.

Offline RL (also called batch RL or fully off-policy RL) relies solely on a previously collected dataset without further interaction. It provides a way to utilize previously collected datasets — from previous RL experiments, from human demonstrations, and from hand-engineered exploration strategies — in order to automatically learn decision-making strategies. In principle, while off-policy RL algorithms can be used in the offline setting (fully off-policy), they are generally only successful when used with active environment interaction — without receiving this direct feedback, they often exhibit undesirable performance in practice. Consequently, while offline RL has enormous potential, that potential cannot be reached without resolving significant algorithmic challenges.

In “Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems”, we provide a comprehensive tutorial on approaches for tackling the challenges of offline RL and discuss the many issues that remain. To address these issues, we have designed and released an open-source benchmarking framework, Datasets for Deep Data-Driven Reinforcement Learning (D4RL), as well as a new, simple, and highly effective offline RL algorithm, called conservative Q-learning (CQL).

Benchmarks for Offline RL
In order to understand the capabilities of current approaches and to guide future progress, it is first necessary to have effective benchmarks. A common choice in prior work was to simply use data generated by a successful online RL run. However, while simple, this data collection approach is artificial because it involves training an online RL agent which is prohibitive in many real-world settings as we discussed previously. One wishes to learn a policy that is better than the current best from diverse data sources that provides good coverage of the task. For example, one might have data collected from a hand-designed controller of a robot arm, and use offline RL to train an improved controller. To enable progress in this field under realistic settings, one needs a benchmark suite that accurately reflects these settings, while being simple and accessible enough to enable rapid experimentation.

D4RL provides standardized environments, datasets and evaluation protocols, as well as reference scores for recent algorithms to help accomplish this. This is a “batteries-included” resource, making it ideal for anyone to jump in and get started with minimal fuss.

Environments in D4RL

The key design goal for D4RL was to develop tasks that reflect both real-world dataset challenges as well as real-world applications. Previous datasets used data collected either from random agents or agents trained with RL. Instead, by thinking through potential applications in autonomous driving, robotics, and other domains, we considered how real-world applications of offline RL might require handling of data generated from human demonstrations or hard-coded controllers, data collected from heterogeneous sources, and data collected by agents with a variety of different goals.

Aside from the widely used MuJoCo locomotion tasks, D4RL includes datasets for more complex tasks. The Adroit domain, which requires manipulating a realistic robotic hand to use a hammer, for example, illustrates the challenges of working with limited human demonstrations, without which these tasks are extremely challenging. Previous work found that existing datasets could not distinguish between competing methods, whereas the Adroit domain reveals clear deficiencies between them.

Another common scenario for real-world tasks is one in which the dataset used for training is collected from agents performing a wide range of other activities that are related to, but not specifically targeted towards, the task of interest. For example, data from human drivers may illustrate how to drive a car well, but do not necessarily show how to reach a specific desired destination. In this case, one might like offline RL methods to “stitch” together parts of routes in the driving dataset to accomplish a task that was not actually seen in the data (i.e., navigation). As an illustrative example, given paths labeled “A” and “B” in the picture below, offline RL should be able to “remix” them to produce path C.

Having only observed paths A and B, they can be combined to form a shortest path (C).

We constructed a series of increasingly difficult tasks to exercise this “stitching” ability. The maze environments, shown below, require two robots (a simple ball or an “Ant” robot) to navigate to locations in a series of mazes.

Maze navigation environments in D4RL, which require “stitching” parts of paths to accomplish new navigational goals that were not seen in the dataset.

A more complex “stitching” scenario is provided by the Franka kitchen domain (based on the Adept environment), where demonstrations from humans using a VR interface comprise a multi-task dataset, and offline RL methods must again “remix” this data.

The “Franka kitchen” domain requires using data from human demonstrators performing a variety of different tasks in a simulated kitchen.

Finally, D4RL includes two tasks that are meant to more accurately reflect potential realistic applications of offline RL, both based on existing driving simulators. One is a first-person driving dataset that utilizes the widely used CARLA simulator developed at Intel, which provides photo-realistic images in realistic driving domains, and the other is a dataset from the Flow traffic control simulator (from UC Berkeley), which requires controlling autonomous vehicles to facilitate effective traffic flow.

D4RL includes datasets based on existing realistic simulators for driving with CARLA (left) and traffic management with Flow (right).

We have packaged these tasks and standardized datasets into an easy-to-use Python package to accelerate research. Furthermore, we provide benchmark numbers for all tasks using relevant prior methods (BC, SAC, BEAR, BRAC, AWR, BCQ), in order to baseline new approaches. We are not the first to propose a benchmark for offline RL: a number of prior works have proposed simple datasets based on running RL algorithms, and several more recent works have proposed datasets with image observations and other features. However, we believe that the more realistic dataset composition in D4RL makes it an effective way to drive progress in the field.

An Improved Algorithm for Offline RL
As we developed the benchmark tasks, we found that existing methods could not solve the more challenging tasks. The central challenge arises from a distributional shift: in order to improve over the historical data, offline RL algorithms must learn to make decisions that differ from the decisions taken in the dataset. However, this can lead to problems when the consequences of a seemingly good decision cannot be deduced from the data — if no agent has taken this particular turn in the maze, how does one know if it leads to the goal or not? Without handling this distributional shift problem, offline RL methods can extrapolate erroneously, making over-optimistic conclusions about the outcomes of rarely seen actions. Contrast this with the online setting, where reward bonuses modeled after curiosity and surprise optimistically bias the agent to explore all potentially rewarding paths. Because the agent receives interactive feedback, if the action turns out to be unrewarding, then it can simply avoid the path in the future.

To address this, we developed conservative Q-learning (CQL), an offline RL algorithm designed to guard against overestimation while avoiding explicit construction of a separate behavior model and without using importance weights. While standard Q-learning (and actor-critic) methods bootstrap from previous estimates, CQL is unique in that it is fundamentally a pessimistic algorithm: it assumes that if a good outcome was not seen for a given action, that action is likely to not be a good one. The central idea of CQL is to learn a lower bound on the policy’s expected return (called the Q-function), instead of learning to approximate the expected return. If we then optimize our policy under this conservative Q-function, we can be confident that its value is no lower than this estimate, preventing errors from overestimation.

We found that CQL attains state-of-the-art results on many of the harder D4RL tasks: CQL outperformed other approaches on the AntMaze, Kitchen tasks, and 6 out of 8 Adroit tasks. In particular, on the AntMaze tasks, which require navigating through a maze with an “Ant” robot, CQL is often the only algorithm that is able to learn non-trivial policies. CQL also performs well on other tasks, including Atari games. On the Atari tasks from Agarwal et al., CQL outperforms prior methods when data is limited (“1%” dataset). Moreover, CQL is simple to implement on top of existing algorithms (e.g., QR-DQN and SAC), without training additional neural networks.

Performance of CQL on Atari games with the 1% dataset from Agarwal et al.

Future Thoughts
We are excited about the fast-moving field of offline RL. While we took a first step towards a standard benchmark, there is clearly still room for improvement. We expect that as algorithms improve, we will need to reevaluate the tasks in the benchmark and develop more challenging tasks. We look forward to working with the community to evolve the benchmark and evaluation protocols. Together, we can bring the rich promises of offline RL to real-world applications.

Acknowledgements
This work was carried out in collaboration with UC Berkeley PhD students Aviral Kumar, Justin Fu, and Aurick Zhou, with contributions from Ofir Nachum from Google Research.

Read More

Posted by Lu Jiang, Senior Research Scientist and Weilong Yang, Senior Staff Software Engineer, Google Research

The success of deep neural networks depends on access to high-quality labeled training data, as the presence of label errors (label noise) in training data can greatly reduce the accuracy of models on clean test data. Unfortunately, large training datasets almost always contain examples with inaccurate or incorrect labels. This leads to a paradox: on one hand, large datasets are necessary to train better deep networks, while on the other hand, deep networks tend to memorize training label noise, resulting in poorer model performance in practice.

The research community has recognized the importance of this problem, introducing works attempting to understand noisy training labels, e.g., by Arpit et al., as well as mitigation strategies, such as MentorNet or co-teaching, to overcome them. Controlled experiments play a crucial role in understanding noisy labels by studying the impact of the noise level — the percentage of examples with incorrect labels in the dataset — on model performance. However, current experiments have only been performed on synthetic labels, in which noisy examples have randomly assigned labels, not real-world label noise, which follows a different noise distribution. Such studies may then result in very different or even contradictory findings about noisy labels compared to practical experience. In addition, methods that perform well on synthetic noise may not work as well on real-world noisy labels.

In “Beyond Synthetic Noise: Deep Learning on Controlled Noisy Labels”, published at ICML 2020, we make three contributions towards better understanding deep learning on non-synthetic noisy labels. First, we establish the first controlled dataset and benchmark of realistic, real-world label noise sourced from the web (i.e., web label noise). Second, we propose a simple but highly effective method to overcome both synthetic and real-world noisy labels. Finally, we conduct the largest study to date that compares synthetic and web label noise across a wide variety of settings.

Properties of Synthetic vs Real-World (Web) Label Noise
There are a number of differences between the distribution of images with synthetic versus real-world (web) label noise. First, images with web label noise tend to be more consistent, visually or semantically, with the true positive images. Second, synthetic label noise is at class-level (all examples in the same class are equally noisy), whereas real-world label noise is at instance-level (certain images are more likely to be mislabelled than others, regardless of the associated class). For example, images of “Honda Civic” and “Honda Accord” are more often confused when the images are taken from the side than when the vehicles are imaged from the front. Third, images with real-world label noise come from an open class vocabulary that may not overlap with the class vocabulary of a specific dataset. For example, the web noisy images of “ladybug” include classes such as “fly” and other bugs that are not included in the class list of the dataset being used. The benchmark for controlled label noise will help provide better quantitative understanding of the differences between synthetic and real-world web label noise.

Benchmark for Controlled Label Noise from the Web
The benchmark in this work is built on two public datasets: Mini-ImageNet, for coarse-grained image classification, and Stanford Cars, for fine-grained image classification. We gradually replace clean images in these datasets with incorrectly labeled images gathered from the web, following standard methods for the construction of synthetic datasets.

To do this, we collect images from the web using the class name (e.g., “ladybug”) as a keyword — an automatic approach to collect noisy labeled images from the web without manual annotations. Each retrieved image is then examined by 3-5 annotators using Google Cloud Labeling Service who identify whether or not the web label given is correct, yielding nearly 213k annotated images. We use these web images with incorrect labels to replace a percentage of the clean training images in the original Mini-ImageNet and Stanford Cars datasets. We create 10 different datasets with progressively higher levels of label noise (from 0% clean data to 80% data with erroneous labels). The datasets have been open-sourced at our Controlled Noisy Web Labels website.

Comparison of synthetic label noise and web label noise. From left to right, columns are true positive images in the Mini-ImageNet or Stanford Cars dataset, images with incorrect synthetic labels, and images with incorrect web labels (collected in the present work).

MentorMix: A Simple Robust Learning Method
Given a dataset of some unknown noise level, our goal is to train a robust model that can generalize well on the clean test data. Building on two existing techniques, MentorNet and Mixup, we introduce a simple yet effective method called MentorMix, which works with a given model of interest to overcome both synthetic and real-world noisy labels.

MentorMix is an iterative approach that comprises four steps: weight, sample, mixup, and weight again. In the first step, a weight is computed for every example in a mini-batch by a MentorNet network, which can be tailored to the task at hand, and the weights are normalized into a distribution. In practice, the goal is to assign high weights for correctly labeled examples and zero weights for incorrectly labeled examples. In reality, we don’t know which are correct and which are incorrect, so MentorNet weights are based on approximations. In the example here, MentorNet uses the StudentNet training loss to determine the weights in the distribution. 

Next, for each example, we use importance sampling to select another example in the same mini-batch according to the distribution. As examples with higher weights tend to have the correct label, they are favored in the sampling procedure. We then use Mixup to mix the original and sampled examples to regularize the model prediction between noisy training examples. Finally, we may compute another weight for the mixed example to scale the final loss. The impact of this second weighting strategy becomes more pronounced for high noise levels. 

Conceptually, the above steps implement a new robust loss, which turns out to be more resilient to noisy training labels. More discussion on this topic can be found in our paper. The animation below illustrates the four key steps in MentorMix, where StudentNet is the model to be trained on noisy labeled data. We employ a very simple version of MentorNet, as described by Jiang et al., to compute the weight for each example.

Illustration of four steps in the MentorMix method: weight, sample, mixup, and weight again.

Evaluation
We evaluate MentorMix on five datasets including CIFAR 10/100 with synthetic label noise, and WebVision 1.0, a large dataset of 2.2 million images with real-world noisy labels. MentorMix consistently yields improved results on the CIFAR 10/100 datasets and achieves the best published result on the WebVision dataset, improving the previous best method by a significant ~3% in terms of the top-1 classification accuracy on the ImageNet ILSVRC12 validation set.

Our model is trained only on the WebVision 2.2 million noisy training sample and is tested on the ImageNet ILSVRC12 validation set. The baseline models reported are (Lee et al. 2018), (MentorNet 2018), and (Guo et al. 2018).

New Findings on Noisy Labels from the Web
This work represents the largest study to date into understanding deep neural networks trained on noisy labels. We propose three new findings on web label noise:

• Deep neural networks generalize much better on web label noise

  While it is well known that deep neural networks generalize poorly on synthetic label noise, our results suggest that deep neural networks generalize much better on web label noise. For example, the classification accuracy of a network trained on the Stanford Cars dataset using the 60% web label noise level is 0.66, much higher than that for the same network trained at the same 60% level of synthetic noise, which achieves only 0.09. This pattern is consistent across our two datasets using both fine-tuning and training from scratch.

• Deep neural networks may NOT learn patterns first when trained on web label noise

  Our common understanding is that deep neural networks learn patterns first — an interesting property in which DNNs are able to automatically capture generalizable “patterns” in the early training stage before memorizing noisy training labels. Because of this, early stopping is commonly used for training on noisy data. However, our results suggest deep neural networks may not learn patterns first when trained using datasets with web label noise, at least for the fine-grained classification task, suggesting that early stopping may not be effective on real-world label noise from the web.

• ImageNet architectures generalize on noisy training labels when the networks are fine-tuned

  Kornblith et al. (2019) found that fine-tuning more advanced architectures trained on ImageNet tend to perform better on downstream tasks that have clean training labels. Our results extend this finding to noisy training data, showing that a better pre-trained architecture that exhibits better performance when pre-trained on ImageNet is likely to perform better even when it is fine-tuned on noisy training labels.

Summary
Based on our findings, we have the following practical recommendations for training deep neural networks on noisy data.

1. A simple way to deal with noisy labels is to fine-tune a model that is pre-trained on clean datasets, like ImageNet. The better the pre-trained model is, the better it may generalize on downstream noisy training tasks.
2. Early stopping may not be effective on the real-world label noise from the web.
3. Methods that perform well on synthetic noise may not work as well on the real-world noisy labels from the web.
4. The label noise from the web appears to be less harmful, yet it is more difficult for our current robust learning methods to tackle. This encourages more future research to be carried out on controlled real-world label noise.
5. The proposed MentorMix can better overcome both synthetic and real-world noisy labels.

The code of MentorMix is available on GitHub, the datasets are on our Dataset Website.

Aknowledgements
This research was conducted by Lu Jiang, Di Huang, Mason Liu, and Weilong Yang. We’d like to thank Boqing Gong and Fei Sha for constructive feedback. Additional thanks go to the leadership Andrew Moore for supporting our data labeling effort, along with Tomas Izo and Rahul Sukthankar for help in releasing the dataset.

Read More

Read More

Posted by Yinfei Yang and Fangxiaoyu Feng, Software Engineers, Google Research

A multilingual embedding model is a powerful tool that encodes text from different languages into a shared embedding space, enabling it to be applied to a range of downstream tasks, like text classification, clustering, and others, while also leveraging semantic information for language understanding. Existing approaches for generating such embeddings, like LASER or m~USE, rely on parallel data, mapping a sentence from one language directly to another language in order to encourage consistency between the sentence embeddings. While these existing multilingual approaches yield good overall performance across a number of languages, they often underperform on high-resource languages compared to dedicated bilingual models, which can leverage approaches like translation ranking tasks with translation pairs as training data to obtain more closely aligned representations. Further, due to limited model capacity and the often poor quality of training data for low-resource languages, it can be difficult to extend multilingual models to support a larger number of languages while maintaining good performance.

Illustration of a multilingual embedding space.

Recent efforts to improve language models include the development of masked language model (MLM) pre-training, such as that used by BERT, ALBERT and RoBERTa. This approach has led to exceptional gains across a wide range of languages and a variety of natural language processing tasks since it only requires monolingual text. In addition, MLM pre-training has been extended to the multilingual setting by modifying MLM training to include concatenated translation pairs, known as translation language modeling (TLM), or by simply introducing pre-training data from multiple languages. However, while the internal model representations learned during MLM and TLM training are helpful when fine-tuning on downstream tasks, without a sentence level objective, they do not directly produce sentence embeddings, which are critical for translation tasks.

In “Language-agnostic BERT Sentence Embedding”, we present a multilingual BERT embedding model, called LaBSE, that produces language-agnostic cross-lingual sentence embeddings for 109 languages. The model is trained on 17 billion monolingual sentences and 6 billion bilingual sentence pairs using MLM and TLM pre-training, resulting in a model that is effective even on low-resource languages for which there is no data available during training. Further, the model establishes a new state of the art on multiple parallel text (a.k.a. bitext) retrieval tasks. We have released the pre-trained model to the community through tfhub, which includes modules that can be used as-is or can be fine-tuned using domain-specific data.

The collection of the training data for 109 supported languages

The Model
In previous work, we proposed the use of a translation ranking task to learn a multilingual sentence embedding space. This approach tasks the model with ranking the true translation over a collection of sentences in the target language, given a sentence in the source language. The translation ranking task is trained using a dual encoder architecture with a shared transformer encoder. The resulting bilingual models achieved state-of-the-art performance on multiple parallel text retrieval tasks (including United Nations and BUCC). However, the model suffered when the bi-lingual models were extended to support multiple languages (16 languages, in our test case) due to limitations in model capacity, vocabulary coverage, training data quality and more.

Translation ranking task. Given a sentence in a given source language, the task is to find the true translation over a collection of sentences in the target language.

For LaBSE, we leverage recent advances on language model pre-training, including MLM and TLM, on a BERT-like architecture and follow this with fine-tuning on a translation ranking task. A 12-layer transformer with a 500k token vocabulary pre-trained using MLM and TLM on 109 languages is used to increase the model and vocabulary coverage. The resulting LaBSE model offers extended support to 109 languages in a single model.

The dual encoder architecture, in which the source and target text are encoded using a shared transformer embedding network separately. The translation ranking task is applied, forcing the text that paraphrases each other to have similar representations. The transformer embedding network is initialized from a BERT checkpoint trained on MLM and TLM tasks.

Performance on Cross-lingual Text Retrieval
We evaluate the proposed model using the Tatoeba corpus, a dataset consisting of up to 1,000 English-aligned sentence pairs for 112 languages. For more than 30 of the languages in the dataset, the model has no training data. The model is tasked with finding the nearest neighbor translation for a given sentence, which it calculates using the cosine distance.

To understand the performance of the model for languages at the head or tail of the training data distribution, we divide the set of languages into several groups and compute the average accuracy for each set. The first 14-language group is selected from the languages supported by m~USE, which cover the languages from the head of the distribution (head languages). We also evaluate a second language group composed of 36 languages from the XTREME benchmark. The third 82-language group, selected from the languages covered by the LASER training data, includes many languages from the tail of the distribution (tail languages). Finally, we compute the average accuracy for all languages.

The table below presents the average accuracy achieved by LaBSE, compared to the m~USE and LASER models, for each language group. As expected, all models perform strongly on the 14-language group that covers most head languages. With more languages included, the averaged accuracy for both LASER and LaBSE declines. However, the reduction in accuracy from the LaBSE model with increasing numbers of languages is much less significant, outperforming LASER significantly, particularly when the full distribution of 112 languages is included (83.7% accuracy vs. 65.5%).

Support to Unsupported Languages
The average performance of all languages included in Tatoeba is very promising. Interestingly, LaBSE even performs relatively well for many of the 30+ Tatoeba languages for which it has no training data (see below). For one third of these languages the LaBSE accuracy is higher than 75% and only 8 have accuracy lower than 25%, indicating very strong transfer performance to languages without training data. Such positive language transfer is only possible due to the massively multilingual nature of LaBSE.

LaBSE accuracy for the subset of Tatoeba languages (represented with ISO 639-1/639-2 codes) for which there was no training data.

Mining Parallel Text from WebLaBSE can be used for mining parallel text (bi-text) from web-scale data. For example, we applied LaBSE to CommonCrawl, a large-scale monolingual corpus, to process 560 million Chinese and 330 million German sentences for the extraction of parallel text. Each Chinese and German sentence pair is encoded using the LaBSE model and then the encoded embedding is used to find a potential translation from a pool of 7.7 billion English sentences pre-processed and encoded by the model. An approximate nearest neighbor search is employed to quickly search through the high-dimensional sentence embeddings. After a simple filtering, the model returns 261M and 104M potential parallel pairs for English-Chinese and English-German, respectively. The trained NMT model using the mined data reaches BLEU scores of 35.7 and 27.2 on the WMT translation tasks (wmt17 for English-to-Chinese and wmt14 for English-to-German). The performance is only a few points away from current state-of-art-models trained on high quality parallel data.

ConclusionWe’re excited to share this research, and the model, with the community. The pre-trained model is released at tfhub to support further research on this direction and possible downstream applications. We also believe that what we’re showing here is just the beginning, and there are more important research problems to be addressed, such as building better models to support all languages.

AcknowledgementsThe core team includes Wei Wang, Naveen Arivazhagan, Daniel Cer. We would like to thank the Google Research Language team, along with our partners in other Google groups for their feedback and suggestions. Special thanks goes to Sidharth Mudgal, and Jax Law for help with data processing; as well as Jialu Liu, Tianqi Liu, Chen Chen, and Anosh Raj for help on BERT pre-training.

Read More

Posted by Valentin Bazarevsky and Ivan Grishchenko, Research Engineers, Google Research

Pose estimation from video plays a critical role enabling the overlay of digital content and information on top of the physical world in augmented reality, sign language recognition, full-body gesture control, and even quantifying physical exercises, where it can form the basis for yoga, dance, and fitness applications. Pose estimation for fitness applications is particularly challenging due to the wide variety of possible poses (e.g., hundreds of yoga asanas), numerous degrees of freedom, occlusions (e.g. the body or other objects occlude limbs as seen from the camera), and a variety of appearances or outfits.

Today we are announcing the release of a new approach to human body pose perception, BlazePose, which we presented at the CV4ARVR workshop at CVPR 2020. Our approach provides human pose tracking by employing machine learning (ML) to infer 33, 2D landmarks of a body from a single frame. In contrast to current pose models based on the standard COCO topology, BlazePose accurately localizes more keypoints, making it uniquely suited for fitness applications. In addition, current state-of-the-art approaches rely primarily on powerful desktop environments for inference, whereas our method achieves real-time performance on mobile phones with CPU inference. If one leverages GPU inference, BlazePose achieves super-real-time performance, enabling it to run subsequent ML models, like face or hand tracking.

Topology
The current standard for human body pose is the COCO topology, which consists of 17 landmarks across the torso, arms, legs, and face. However, the COCO keypoints only localize to the ankle and wrist points, lacking scale and orientation information for hands and feet, which is vital for practical applications like fitness and dance. The inclusion of more keypoints is crucial for the subsequent application of domain-specific pose estimation models, like those for hands, face, or feet.

With BlazePose, we present a new topology of 33 human body keypoints, which is a superset of COCO, BlazeFace and BlazePalm topologies. This allows us to determine body semantics from pose prediction alone that is consistent with face and hand models.

BlazePose 33 keypoint topology as COCO (colored with green) superset

Overview: An ML Pipeline for Pose Tracking
For pose estimation, we utilize our proven two-step detector-tracker ML pipeline. Using a detector, this pipeline first locates the pose region-of-interest (ROI) within the frame. The tracker subsequently predicts all 33 pose keypoints from this ROI. Note that for video use cases, the detector is run only on the first frame. For subsequent frames we derive the ROI from the previous frame’s pose keypoints as discussed below.

Human pose estimation pipeline overview.

Pose Detection by extending BlazeFace
For real-time performance of the full ML pipeline consisting of pose detection and tracking models, each component must be very fast, using only a few milliseconds per frame. To accomplish this, we observe that the strongest signal to the neural network about the position of the torso is the person’s face (due to its high-contrast features and comparably small variations in appearance). Therefore, we achieve a fast and lightweight pose detector by making the strong (yet for many mobile and web applications valid) assumption that the head should be visible for our single-person use case.

Consequently, we trained a face detector, inspired by our sub-millisecond BlazeFace model, as a proxy for a pose detector. Note, this model only detects the location of a person within the frame and can not be used to identify individuals. In contrast to the Face Mesh and MediaPipe Hand tracking pipelines, where we derive the ROI from predicted keypoints, for the human pose tracking we explicitly predict two additional virtual keypoints that firmly describe the human body center, rotation and scale as a circle. Inspired by Leonardo’s Vitruvian man, we predict the midpoint of a person’s hips, the radius of a circle circumscribing the whole person, and the incline angle of the line connecting the shoulder and hip midpoints. This results in consistent tracking even for very complicated cases, like specific yoga asanas. The figure below illustrates the approach.

Vitruvian man aligned via two virtual keypoints predicted by our BlazePose detector in addition to the face bounding box

Tracking Model
The pose estimation component of the pipeline predicts the location of all 33 person keypoints with three degrees of freedom each (x, y location and visibility) plus the two virtual alignment keypoints described above. Unlike current approaches that employ compute-intensive heatmap prediction, our model uses a regression approach that is supervised by a combined heat map/offset prediction of all keypoints, as shown below.

Tracking network architecture: regression with heatmap supervision

Specifically, during training we first employ a heatmap and offset loss to train the center and left tower of the network. We then remove the heatmap output and train the regression encoder (right tower), thus, effectively using the heatmap to supervise a lightweight embedding.

The table below shows an ablation study of the model quality resulting from different training strategies. As an evaluation metric, we use the Percent of Correct Points with 20% tolerance (PCK@0.2) (where we assume the point to be detected correctly if the 2D Euclidean error is smaller than 20% of the corresponding person’s torso size). To obtain a human baseline, we asked annotators to annotate several samples redundantly and obtained an average PCK@0.2 of 97.2. The training and validation have been done on a geo-diverse dataset of various poses, sampled uniformly.

To cover a wide range of customer hardware, we present two pose tracking models: lite and full, which are differentiated in the balance of speed versus quality. For performance evaluation on CPU, we use XNNPACK; for mobile GPUs, we use the TFLite GPU backend.

Applications
Based on human pose, we can build a variety of applications, like fitness or yoga trackers. As an example, we present squats and push up counters, which can automatically count user statistics, or verify the quality of performed exercises. Such use cases can be implemented either using an additional classifier network or even with a simple joint pairwise distance lookup algorithm, which matches the closest pose in normalized pose space.

The number of performed exercises counter based on detected body pose. Left: Squats; Right: Push-Ups

Conclusion
BlazePose will be available to the broader mobile developer community via the Pose detection API in the upcoming release of ML Kit, and we are also releasing a version targeting upper body use cases in MediaPipe running in Android, iOS and Python. Apart from the mobile domain, we preview our web-based in-browser version as well. We hope that providing this human pose perception functionality to the broader research and development community will result in an emergence of creative use cases, stimulating new applications, and new research avenues.

We plan to extend this technology with more robust and stable tracking to an even larger variety of human poses and activities. In the accompanying Model Card, we detail the intended uses, limitations and model fairness to ensure that use of these models aligns with Google’s AI Principles. We believe that publishing this technology can provide an impulse to new creative ideas and applications by the members of the research and developer community at large. We are excited to see what you can build with it!

Acknowledgments
Special thanks to all our team members who worked on the tech with us: Fan Zhang, Artsiom Ablavatski, Yury Kartynnik, Tyler Zhu, Karthik Raveendran, Andrei Vakunov, Andrei Tkachenka, Marat Dukhan, Tyler Mullen, Gregory Karpiak, Suril Shah, Buck Bourdon, Jiuqiang Tang, Ming Guang Yong, Chuo-Ling Chang, Esha Uboweja, Siarhei Kazakou, Andrei Kulik, Matsvei Zhdanovich, and Matthias Grundmann.

Read More

Posted by Ming-Wei Chang and Kelvin Guu, Research Scientists, Google Research

Recent advances in natural language processing have largely built upon the power of unsupervised pre-training, which trains general purpose language representation models using a large amount of text, without human annotations or labels. These pre-trained models, such as BERT and RoBERTa, have been shown to memorize a surprising amount of world knowledge, such as “the birthplace of Francesco Bartolomeo Conti”, “the developer of JDK” and “the owner of Border TV”. While the ability to encode knowledge is especially important for certain natural language processing tasks such as question answering, information retrieval and text generation, these models memorize knowledge implicitly — i.e., world knowledge is captured in an abstract way in the model weights — making it difficult to determine what knowledge has been stored and where it is kept in the model. Furthermore, the storage space, and hence the accuracy of the model, is limited by the size of the network. To capture more world knowledge, the standard practice is to train ever-larger networks, which can be prohibitively slow or expensive.

Instead, what if there was a method for pre-training that could access knowledge explicitly, e.g., by referencing an additional large external text corpus, in order to achieve accurate results without increasing the model size or complexity?  For example, a sentence found in an external document collection, “Francesco Bartolomeo Conti was born in Florence,” could be referenced by the model to determine the birthplace of the musician, rather than relying on the model’s opaque ability to access the knowledge stored in its own parameters. The ability to retrieve text containing explicit knowledge such as this would improve the efficiency of pre-training while enabling the model to perform well on knowledge-intensive tasks without using billions of parameters.

In “REALM: Retrieval-Augmented Language Model Pre-Training”, accepted at the 2020 International Conference on Machine Learning, we share a novel paradigm for language model pre-training, which augments a language representation model with a knowledge retriever, allowing REALM models to retrieve textual world knowledge explicitly from raw text documents, instead of memorizing all the knowledge in the model parameters. We have also open sourced the REALM codebase to demonstrate how one can train the retriever and the language representation jointly.

Background: Pre-training Language Representation Models
To understand how standard language representation models memorize world knowledge, one should first review how these models are pre-trained. Since the invention of BERT, the fill-in-the-blank task, called masked language modeling, has been widely used for pre-training language representation models. Given any text with certain words masked out, the task is to fill back the missing words. An example of this task looks like:

I am so thirsty. I need to __ water.

During pre-training, a model will go over a large number of examples and adjust the parameters in order to predict the missing words (answer: drink, in the above example). Interestingly, the fill-in-the-blank task makes the model memorize certain facts about the world. For example, the knowledge of Einstein’s birthplace is required to fill the missing word in the following example:

Einstein was a __-born scientist. (answer: German)

However, because the world knowledge captured by the model is stored in the model weights, it is abstract, making it difficult to understand what information is stored.

Our Proposal: Retrieval-Augmented Language Representation Model Pre-training
In contrast to standard language representation models, REALM augments the language representation model with a knowledge retriever that first retrieves another piece of text from an external document collection as the supporting knowledge — in our experiments, we use the Wikipedia text corpus — and then feeds this supporting text as well as the original text into a language representation model.

The key intuition of REALM is that a retrieval system should improve the model’s ability to fill in missing words. Therefore, a retrieval that provides more context for filling the missing words should be rewarded. If the retrieved information does not help the model make its predictions, it should be discouraged, making room for better retrievals.

How does one train a knowledge retriever, given that only unlabeled text is available during pre-training? It turns out that one can use the task of filling words to train the knowledge retriever indirectly, without any human annotations. Assume the input of the query is:

We paid twenty __ at the Buckingham Palace gift shop.

Filling the missing word (answer:pounds) in this sentence without retrieval can be tricky, as the model would need to have implicitly stored knowledge of the country in which the Buckingham Palace is located and the associated currency, as well as make the connection between the two. It would be easier for the model to fill in the missing word if it was presented with a passage that explicitly connects some of the necessary knowledge, retrieved from an external corpus.

In this example, the retriever would be rewarded for retrieving the following sentence.

Buckingham Palace is the London residence of the British monarchy.

Since the retrieval step needs to add more context, there may be multiple retrieval targets that could be helpful in filling the missing word, for example, “The official currency of the United Kingdom is the Pound.” The whole process is demonstrated in the next figure:

Computational Challenges for REALM
Scaling REALM pre-training such that models can retrieve knowledge from millions of documents is challenging. In REALM, the selection of the best document is formulated as maximum inner product search (MIPS). To perform retrieval, MIPS models need to first encode all of the documents in the collection, such that each document has a corresponding document vector. When an input arrives, it is encoded as a query vector. In MIPS, given a query, the document in the collection that has the maximum inner product value between its document vector and the query vector is retrieved, as shown in the following figure:

In REALM, we use the ScaNN package to conduct MIPS efficiently, which makes finding the maximum inner product value relatively cheap, given that the document vectors are pre-computed. However, if the model parameters were updated during training, it is typically necessary to re-encode the document vectors for the entire collection of documents. To address the computational challenges, we structure the retriever so that the computation performed for each document can be cached and asynchronously updated. We also found that updating document vectors every 500 training steps, instead of every step, is able to achieve good performance and make training tractable.

Applying REALM to Open-domain Question Answering
We evaluate the effectiveness of REALM by applying it to open-domain question answering (Open-QA), one of the most knowledge-intensive tasks in natural language processing. The goal of the task is to answer questions, such as “What is the angle of the equilateral triangle?”

In standard question answering tasks (e.g., SQuAD or Natural Questions), the supporting document is provided as part of input, so a model only needs to look up the answer in the given document. In Open-QA, there are no given documents, so that Open-QA models need to look up the knowledge by themselves — this makes Open-QA an excellent task to examine the effectiveness of REALM.

The following figure shows the results on the OpenQA version of Natural Question. We mainly compared our results with T5, another approach that trains models without annotated supporting documents. From the figure, one can clearly see that REALM pre-training generates very powerful Open-QA models, and even outperforms the much larger T5 (11B) model by almost 4 points, using only a fraction of the parameters (300M).

Conclusion
The release of REALM has helped drive interest in developing end-to-end retrieval-augmented models, including a recent retrieval-augmented generative model. We look forward to the possibility of extending this line of work in several ways, including 1) applying REALM-like methods to new applications that require knowledge-intensive reasoning and interpretable provenance (beyond Open-QA), and 2) exploring the benefits of retrieving other forms of knowledge, such as images, knowledge graph structures, or even text in other languages. We are also excited to see what the research community does with the open source REALM codebase!

Acknowledgements
This work has been a collaborative effort involving Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat and Ming-Wei Chang.

Read More