Michael J. Wilber

Xuan Yang, Liangzhe Yuan, Kimberly Wilber et al. · deepmind

Dense prediction tasks, such as semantic segmentation, depth estimation, and surface normal prediction, can be easily formulated as per-pixel classification (discrete outputs) or regression (continuous outputs). This per-pixel prediction paradigm has remained popular due to the prevalence of fully convolutional networks. However, on the recent frontier of segmentation task, the community has been witnessing a shift of paradigm from per-pixel prediction to cluster-prediction with the emergence of transformer architectures, particularly the mask transformers, which directly predicts a label for a mask instead of a pixel. Despite this shift, methods based on the per-pixel prediction paradigm still dominate the benchmarks on the other dense prediction tasks that require continuous outputs, such as depth estimation and surface normal prediction. Motivated by the success of DORN and AdaBins in depth estimation, achieved by discretizing the continuous output space, we propose to generalize the cluster-prediction based method to general dense prediction tasks. This allows us to unify dense prediction tasks with the mask transformer framework. Remarkably, the resulting model PolyMaX demonstrates state-of-the-art performance on three benchmarks of NYUD-v2 dataset. We hope our simple yet effective design can inspire more research on exploiting mask transformers for more dense prediction tasks. Code and model will be made available.

Grant Van Horn, Rui Qian, Kimberly Wilber et al. · deepmind

We present a new benchmark dataset, Sapsucker Woods 60 (SSW60), for advancing research on audiovisual fine-grained categorization. While our community has made great strides in fine-grained visual categorization on images, the counterparts in audio and video fine-grained categorization are relatively unexplored. To encourage advancements in this space, we have carefully constructed the SSW60 dataset to enable researchers to experiment with classifying the same set of categories in three different modalities: images, audio, and video. The dataset covers 60 species of birds and is comprised of images from existing datasets, and brand new, expert-curated audio and video datasets. We thoroughly benchmark audiovisual classification performance and modality fusion experiments through the use of state-of-the-art transformer methods. Our findings show that performance of audiovisual fusion methods is better than using exclusively image or audio based methods for the task of video classification. We also present interesting modality transfer experiments, enabled by the unique construction of SSW60 to encompass three different modalities. We hope the SSW60 dataset and accompanying baselines spur research in this fascinating area.

Sagar M. Waghmare, Kimberly Wilber, Dave Hawkey et al. · deepmind

Vision is essential for human navigation. The World Health Organization (WHO) estimates that 43.3 million people were blind in 2020, and this number is projected to reach 61 million by 2050. Modern scene understanding models could empower these people by assisting them with navigation, obstacle avoidance and visual recognition capabilities. The research community needs high quality datasets for both training and evaluation to build these systems. While datasets for autonomous vehicles are abundant, there is a critical gap in datasets tailored for outdoor human navigation. This gap poses a major obstacle to the development of computer vision based Assistive Technologies. To overcome this obstacle, we present SANPO, a large-scale egocentric video dataset designed for dense prediction in outdoor human navigation environments. SANPO contains 701 stereo videos of 30+ seconds captured in diverse real-world outdoor environments across four geographic locations in the USA. Every frame has a high resolution depth map and 112K frames were annotated with temporally consistent dense video panoptic segmentation labels. The dataset also includes 1961 high-quality synthetic videos with pixel accurate depth and panoptic segmentation annotations to balance the noisy real world annotations with the high precision synthetic annotations. SANPO is already publicly available and is being used by mobile applications like Project Guideline to train mobile models that help low-vision users go running outdoors independently. To preserve anonymization during peer review, we will provide a link to our dataset upon acceptance. SANPO is available here: https://google-research-datasets.github.io/sanpo_dataset/

Elijah Cole, Kimberly Wilber, Grant Van Horn et al.

Weakly supervised object localization (WSOL) aims to learn representations that encode object location using only image-level category labels. However, many objects can be labeled at different levels of granularity. Is it an animal, a bird, or a great horned owl? Which image-level labels should we use? In this paper we study the role of label granularity in WSOL. To facilitate this investigation we introduce iNatLoc500, a new large-scale fine-grained benchmark dataset for WSOL. Surprisingly, we find that choosing the right training label granularity provides a much larger performance boost than choosing the best WSOL algorithm. We also show that changing the label granularity can significantly improve data efficiency.

Zheng Xu, Michael Wilber, Chen Fang et al.

We propose a fast feed-forward network for arbitrary style transfer, which can generate stylized image for previously unseen content and style image pairs. Besides the traditional content and style representation based on deep features and statistics for textures, we use adversarial networks to regularize the generation of stylized images. Our adversarial network learns the intrinsic property of image styles from large-scale multi-domain artistic images. The adversarial training is challenging because both the input and output of our generator are diverse multi-domain images. We use a conditional generator that stylized content by shifting the statistics of deep features, and a conditional discriminator based on the coarse category of styles. Moreover, we propose a mask module to spatially decide the stylization level and stabilize adversarial training by avoiding mode collapse. As a side effect, our trained discriminator can be applied to rank and select representative stylized images. We qualitatively and quantitatively evaluate the proposed method, and compare with recent style transfer methods. We release our code and model at https://github.com/nightldj/behance_release.

Marco Fornoni, Chaochao Yan, Liangchen Luo et al.

When interacting with objects through cameras, or pictures, users often have a specific intent. For example, they may want to perform a visual search. With most object detection models relying on image pixels as their sole input, undesired results are not uncommon. Most typically: lack of a high-confidence detection on the object of interest, or detection with a wrong class label. The issue is especially severe when operating capacity-constrained mobile object detectors on-device. In this paper we investigate techniques to modulate mobile detectors to explicitly account for the user intent, expressed as an embedding of a simple query. Compared to standard detectors, query-modulated detectors show superior performance at detecting objects for a given user query. Thanks to large-scale training data synthesized from standard object detection annotations, query-modulated detectors also outperform a specialized referring expression recognition system. Query-modulated detectors can also be trained to simultaneously solve for both localizing a user query and standard detection, even outperforming standard mobile detectors at the canonical COCO task.

Elijah Cole, Xuan Yang, Kimberly Wilber et al.

Recent self-supervised representation learning techniques have largely closed the gap between supervised and unsupervised learning on ImageNet classification. While the particulars of pretraining on ImageNet are now relatively well understood, the field still lacks widely accepted best practices for replicating this success on other datasets. As a first step in this direction, we study contrastive self-supervised learning on four diverse large-scale datasets. By looking through the lenses of data quantity, data domain, data quality, and task granularity, we provide new insights into the necessary conditions for successful self-supervised learning. Our key findings include observations such as: (i) the benefit of additional pretraining data beyond 500k images is modest, (ii) adding pretraining images from another domain does not lead to more general representations, (iii) corrupted pretraining images have a disparate impact on supervised and self-supervised pretraining, and (iv) contrastive learning lags far behind supervised learning on fine-grained visual classification tasks.

Grant Van Horn, Elijah Cole, Sara Beery et al.

Recent progress in self-supervised learning has resulted in models that are capable of extracting rich representations from image collections without requiring any explicit label supervision. However, to date the vast majority of these approaches have restricted themselves to training on standard benchmark datasets such as ImageNet. We argue that fine-grained visual categorization problems, such as plant and animal species classification, provide an informative testbed for self-supervised learning. In order to facilitate progress in this area we present two new natural world visual classification datasets, iNat2021 and NeWT. The former consists of 2.7M images from 10k different species uploaded by users of the citizen science application iNaturalist. We designed the latter, NeWT, in collaboration with domain experts with the aim of benchmarking the performance of representation learning algorithms on a suite of challenging natural world binary classification tasks that go beyond standard species classification. These two new datasets allow us to explore questions related to large-scale representation and transfer learning in the context of fine-grained categories. We provide a comprehensive analysis of feature extractors trained with and without supervision on ImageNet and iNat2021, shedding light on the strengths and weaknesses of different learned features across a diverse set of tasks. We find that features produced by standard supervised methods still outperform those produced by self-supervised approaches such as SimCLR. However, improved self-supervised learning methods are constantly being released and the iNat2021 and NeWT datasets are a valuable resource for tracking their progress.

Srinadh Bhojanapalli, Kimberly Wilber, Andreas Veit et al.

Standard training techniques for neural networks involve multiple sources of randomness, e.g., initialization, mini-batch ordering and in some cases data augmentation. Given that neural networks are heavily over-parameterized in practice, such randomness can cause {\em churn} -- for the same input, disagreements between predictions of the two models independently trained by the same algorithm, contributing to the `reproducibility challenges' in modern machine learning. In this paper, we study this problem of churn, identify factors that cause it, and propose two simple means of mitigating it. We first demonstrate that churn is indeed an issue, even for standard image classification tasks (CIFAR and ImageNet), and study the role of the different sources of training randomness that cause churn. By analyzing the relationship between churn and prediction confidences, we pursue an approach with two components for churn reduction. First, we propose using \emph{minimum entropy regularizers} to increase prediction confidences. Second, \changes{we present a novel variant of co-distillation approach~\citep{anil2018large} to increase model agreement and reduce churn}. We present empirical results showing the effectiveness of both techniques in reducing churn while improving the accuracy of the underlying model.

Andreas Veit, Kimberly Wilber

Modern image retrieval systems increasingly rely on the use of deep neural networks to learn embedding spaces in which distance encodes the relevance between a given query and image. In this setting, existing approaches tend to emphasize one of two properties. Triplet-based methods capture top-$k$ relevancy, where all top-$k$ scoring documents are assumed to be relevant to a given query Pairwise contrastive models capture threshold relevancy, where all documents scoring higher than some threshold are assumed to be relevant. In this paper, we propose Cross-Example Softmax which combines the properties of top-$k$ and threshold relevancy. In each iteration, the proposed loss encourages all queries to be closer to their matching images than all queries are to all non-matching images. This leads to a globally more calibrated similarity metric and makes distance more interpretable as an absolute measure of relevance. We further introduce Cross-Example Negative Mining, in which each pair is compared to the hardest negative comparisons across the entire batch. Empirically, we show in a series of experiments on Conceptual Captions and Flickr30k, that the proposed method effectively improves global calibration and also retrieval performance.

Xiao Ma, Lina Mezghani, Kimberly Wilber et al.

As any savvy online shopper knows, second-hand peer-to-peer marketplaces are filled with images of mixed quality. How does image quality impact marketplace outcomes, and can quality be automatically predicted? In this work, we conducted a large-scale study on the quality of user-generated images in peer-to-peer marketplaces. By gathering a dataset of common second-hand products (~75,000 images) and annotating a subset with human-labeled quality judgments, we were able to model and predict image quality with decent accuracy (~87%). We then conducted two studies focused on understanding the relationship between these image quality scores and two marketplace outcomes: sales and perceived trustworthiness. We show that image quality is associated with higher likelihood that an item will be sold, though other factors such as view count were better predictors of sales. Nonetheless, we show that high quality user-generated images selected by our models outperform stock imagery in eliciting perceptions of trust from users. Our findings can inform the design of future marketplaces and guide potential sellers to take better product images.

Michael J. Wilber, Chen Fang, Hailin Jin et al.

Computer vision systems are designed to work well within the context of everyday photography. However, artists often render the world around them in ways that do not resemble photographs. Artwork produced by people is not constrained to mimic the physical world, making it more challenging for machines to recognize. This work is a step toward teaching machines how to categorize images in ways that are valuable to humans. First, we collect a large-scale dataset of contemporary artwork from Behance, a website containing millions of portfolios from professional and commercial artists. We annotate Behance imagery with rich attribute labels for content, emotions, and artistic media. Furthermore, we carry out baseline experiments to show the value of this dataset for artistic style prediction, for improving the generality of existing object classifiers, and for the study of visual domain adaptation. We believe our Behance Artistic Media dataset will be a good starting point for researchers wishing to study artistic imagery and relevant problems.

Andreas Veit, Michael Wilber, Serge Belongie

In this work we propose a novel interpretation of residual networks showing that they can be seen as a collection of many paths of differing length. Moreover, residual networks seem to enable very deep networks by leveraging only the short paths during training. To support this observation, we rewrite residual networks as an explicit collection of paths. Unlike traditional models, paths through residual networks vary in length. Further, a lesion study reveals that these paths show ensemble-like behavior in the sense that they do not strongly depend on each other. Finally, and most surprising, most paths are shorter than one might expect, and only the short paths are needed during training, as longer paths do not contribute any gradient. For example, most of the gradient in a residual network with 110 layers comes from paths that are only 10-34 layers deep. Our results reveal one of the key characteristics that seem to enable the training of very deep networks: Residual networks avoid the vanishing gradient problem by introducing short paths which can carry gradient throughout the extent of very deep networks.

Michael J. Wilber, Vitaly Shmatikov, Serge Belongie

After decades of study, automatic face detection and recognition systems are now accurate and widespread. Naturally, this means users who wish to avoid automatic recognition are becoming less able to do so. Where do we stand in this cat-and-mouse race? We currently live in a society where everyone carries a camera in their pocket. Many people willfully upload most or all of the pictures they take to social networks which invest heavily in automatic face recognition systems. In this setting, is it still possible for privacy-conscientious users to avoid automatic face detection and recognition? If so, how? Must evasion techniques be obvious to be effective, or are there still simple measures that users can use to protect themselves? In this work, we find ways to evade face detection on Facebook, a representative example of a popular social network that uses automatic face detection to enhance their service. We challenge widely-held beliefs about evading face detection: do our old techniques such as blurring the face region or wearing "privacy glasses" still work? We show that in general, state-of-the-art detectors can often find faces even if the subject wears occluding clothing or even if the uploader damages the photo to prevent faces from being detected.

Andreas Veit, Michael Wilber, Rajan Vaish et al.

When crowdsourcing systems are used in combination with machine inference systems in the real world, they benefit the most when the machine system is deeply integrated with the crowd workers. However, if researchers wish to integrate the crowd with "off-the-shelf" machine classifiers, this deep integration is not always possible. This work explores two strategies to increase accuracy and decrease cost under this setting. First, we show that reordering tasks presented to the human can create a significant accuracy improvement. Further, we show that greedily choosing parameters to maximize machine accuracy is sub-optimal, and joint optimization of the combined system improves performance.

Michael J. Wilber, Iljung S. Kwak, David Kriegman et al.

This paper presents our work on "SNaCK," a low-dimensional concept embedding algorithm that combines human expertise with automatic machine similarity kernels. Both parts are complimentary: human insight can capture relationships that are not apparent from the object's visual similarity and the machine can help relieve the human from having to exhaustively specify many constraints. We show that our SNaCK embeddings are useful in several tasks: distinguishing prime and nonprime numbers on MNIST, discovering labeling mistakes in the Caltech UCSD Birds (CUB) dataset with the help of deep-learned features, creating training datasets for bird classifiers, capturing subjective human taste on a new dataset of 10,000 foods, and qualitatively exploring an unstructured set of pictographic characters. Comparisons with the state-of-the-art in these tasks show that SNaCK produces better concept embeddings that require less human supervision than the leading methods.

Jack Hessel, Nicolas Savva, Michael J. Wilber

We examine the possibility that recent promising results in automatic caption generation are due primarily to language models. By varying image representation quality produced by a convolutional neural network, we find that a state-of-the-art neural captioning algorithm is able to produce quality captions even when provided with surprisingly poor image representations. We replicate this result in a new, fine-grained, transfer learned captioning domain, consisting of 66K recipe image/title pairs. We also provide some experiments regarding the appropriateness of datasets for automatic captioning, and find that having multiple captions per image is beneficial, but not an absolute requirement.

Michael J. Wilber, Iljung S. Kwak, Serge J. Belongie

Similarity comparisons of the form "Is object a more similar to b than to c?" are useful for computer vision and machine learning applications. Unfortunately, an embedding of $n$ points is specified by $n^3$ triplets, making collecting every triplet an expensive task. In noticing this difficulty, other researchers have investigated more intelligent triplet sampling techniques, but they do not study their effectiveness or their potential drawbacks. Although it is important to reduce the number of collected triplets, it is also important to understand how best to display a triplet collection task to a user. In this work we explore an alternative display for collecting triplets and analyze the monetary cost and speed of the display. We propose best practices for creating cost effective human intelligence tasks for collecting triplets. We show that rather than changing the sampling algorithm, simple changes to the crowdsourcing UI can lead to much higher quality embeddings. We also provide a dataset as well as the labels collected from crowd workers.

Walter J. Scheirer, Michael J. Wilber, Michael Eckmann et al.

Recognition is the fundamental task of visual cognition, yet how to formalize the general recognition problem for computer vision remains an open issue. The problem is sometimes reduced to the simplest case of recognizing matching pairs, often structured to allow for metric constraints. However, visual recognition is broader than just pair matching -- especially when we consider multi-class training data and large sets of features in a learning context. What we learn and how we learn it has important implications for effective algorithms. In this paper, we reconsider the assumption of recognition as a pair matching test, and introduce a new formal definition that captures the broader context of the problem. Through a meta-analysis and an experimental assessment of the top algorithms on popular data sets, we gain a sense of how often metric properties are violated by good recognition algorithms. By studying these violations, useful insights come to light: we make the case that locally metric algorithms should leverage outside information to solve the general recognition problem.