Sarah Adel Bargal

Nataniel Ruiz, Sarah Adel Bargal, Cihang Xie et al.

Modern deep neural networks tend to be evaluated on static test sets. One shortcoming of this is the fact that these deep neural networks cannot be easily evaluated for robustness issues with respect to specific scene variations. For example, it is hard to study the robustness of these networks to variations of object scale, object pose, scene lighting and 3D occlusions. The main reason is that collecting real datasets with fine-grained naturalistic variations of sufficient scale can be extremely time-consuming and expensive. In this work, we present Counterfactual Simulation Testing, a counterfactual framework that allows us to study the robustness of neural networks with respect to some of these naturalistic variations by building realistic synthetic scenes that allow us to ask counterfactual questions to the models, ultimately providing answers to questions such as "Would your classification still be correct if the object were viewed from the top?" or "Would your classification still be correct if the object were partially occluded by another object?". Our method allows for a fair comparison of the robustness of recently released, state-of-the-art Convolutional Neural Networks and Vision Transformers, with respect to these naturalistic variations. We find evidence that ConvNext is more robust to pose and scale variations than Swin, that ConvNext generalizes better to our simulated domain and that Swin handles partial occlusion better than ConvNext. We also find that robustness for all networks improves with network scale and with data scale and variety. We release the Naturalistic Variation Object Dataset (NVD), a large simulated dataset of 272k images of everyday objects with naturalistic variations such as object pose, scale, viewpoint, lighting and occlusions. Project page: https://counterfactualsimulation.github.io

Dina Bashkirova, Samarth Mishra, Diala Lteif et al.

Label-efficient and reliable semantic segmentation is essential for many real-life applications, especially for industrial settings with high visual diversity, such as waste sorting. In industrial waste sorting, one of the biggest challenges is the extreme diversity of the input stream depending on factors like the location of the sorting facility, the equipment available in the facility, and the time of year, all of which significantly impact the composition and visual appearance of the waste stream. These changes in the data are called ``visual domains'', and label-efficient adaptation of models to such domains is needed for successful semantic segmentation of industrial waste. To test the abilities of computer vision models on this task, we present the VisDA 2022 Challenge on Domain Adaptation for Industrial Waste Sorting. Our challenge incorporates a fully-annotated waste sorting dataset, ZeroWaste, collected from two real material recovery facilities in different locations and seasons, as well as a novel procedurally generated synthetic waste sorting dataset, SynthWaste. In this competition, we aim to answer two questions: 1) can we leverage domain adaptation techniques to minimize the domain gap? and 2) can synthetic data augmentation improve performance on this task and help adapt to changing data distributions? The results of the competition show that industrial waste detection poses a real domain adaptation problem, that domain generalization techniques such as augmentations, ensembling, etc., improve the overall performance on the unlabeled target domain examples, and that leveraging synthetic data effectively remains an open problem. See https://ai.bu.edu/visda-2022/

Ana-Maria Cretu, Klim Kireev, Amro Abdalla et al.

We evaluate the effectiveness of filtering child images from training datasets of text-to-image models to prevent model misuse to create child sexual abuse material (CSAM). First, we capture the complexity of preventing CSAM generation using a game-based security definition. Second, we show that current detection methods cannot remove all children from a dataset. Third, using an ethical proxy for CSAM (a child wearing glasses), we show that even when only a small percentage of child images are left in the training dataset after filtering, there exist prompting strategies that generate a child wearing glasses using only a few more queries than when the model is trained on the unfiltered data. Fine-tuning the filtered model on child images further reduces the additional query overhead. We also show that re-introducing a concept is possible via fine-tuning even if filtering is perfect. Our results show that current child filtering methods offer limited protection to closed-weight models and no protection to open-weight models, while reducing the generality of the model by hindering the generation of child-related concepts or changing their representation. We conclude by outlining challenges in conducting evaluations that establish robust evidence on the impact of concept filtering defenses for CSAM.

Quanfu Fan, Donghyun Kim, Chun-Fu et al.

In this paper, we provide a deep analysis of temporal modeling for action recognition, an important but underexplored problem in the literature. We first propose a new approach to quantify the temporal relationships between frames captured by CNN-based action models based on layer-wise relevance propagation. We then conduct comprehensive experiments and in-depth analysis to provide a better understanding of how temporal modeling is affected by various factors such as dataset, network architecture, and input frames. With this, we further study some important questions for action recognition that lead to interesting findings. Our analysis shows that there is no strong correlation between temporal relevance and model performance; and action models tend to capture local temporal information, but less long-range dependencies. Our codes and models will be publicly available.

Ariel N. Lee, Sarah Adel Bargal, Janavi Kasera et al.

Vision transformers (ViTs) have significantly changed the computer vision landscape and have periodically exhibited superior performance in vision tasks compared to convolutional neural networks (CNNs). Although the jury is still out on which model type is superior, each has unique inductive biases that shape their learning and generalization performance. For example, ViTs have interesting properties with respect to early layer non-local feature dependence, as well as self-attention mechanisms which enhance learning flexibility, enabling them to ignore out-of-context image information more effectively. We hypothesize that this power to ignore out-of-context information (which we name $\textit{patch selectivity}$), while integrating in-context information in a non-local manner in early layers, allows ViTs to more easily handle occlusion. In this study, our aim is to see whether we can have CNNs $\textit{simulate}$ this ability of patch selectivity by effectively hardwiring this inductive bias using Patch Mixing data augmentation, which consists of inserting patches from another image onto a training image and interpolating labels between the two image classes. Specifically, we use Patch Mixing to train state-of-the-art ViTs and CNNs, assessing its impact on their ability to ignore out-of-context patches and handle natural occlusions. We find that ViTs do not improve nor degrade when trained using Patch Mixing, but CNNs acquire new capabilities to ignore out-of-context information and improve on occlusion benchmarks, leaving us to conclude that this training method is a way of simulating in CNNs the abilities that ViTs already possess. We will release our Patch Mixing implementation and proposed datasets for public use. Project page: https://arielnlee.github.io/PatchMixing/

Dina Bashkirova, Arijit Ray, Rupayan Mallick et al.

Professional artists, photographers, and other visual content creators use object relighting to establish their photo's desired effect. Unfortunately, manual tools that allow relighting have a steep learning curve and are difficult to master. Although generative editing methods now enable some forms of image editing, relighting is still beyond today's capabilities; existing methods struggle to keep other aspects of the image -- colors, shapes, and textures -- consistent after the edit. We propose Lasagna, a method that enables intuitive text-guided relighting control. Lasagna learns a lighting prior by using score distillation sampling to distill the prior of a diffusion model, which has been finetuned on synthetic relighting data. To train Lasagna, we curate a new synthetic dataset ReLiT, which contains 3D object assets re-lit from multiple light source locations. Despite training on synthetic images, quantitative results show that Lasagna relights real-world images while preserving other aspects of the input image, outperforming state-of-the-art text-guided image editing methods. Lasagna enables realistic and controlled results on natural images and digital art pieces and is preferred by humans over other methods in over 91% of cases. Finally, we demonstrate the versatility of our learning objective by extending it to allow colorization, another form of image editing.

Dan DeGenaro, Xin Li, Obed Amo et al.

We introduce FLASH-MAX, a shallow, exact-by-construction neural network architecture for predicting homogeneous electromagnetic fields from sparse pointwise observations. Each hidden neuron represents a separate exact solution to Maxwell's equations, so that the network satisfies the governing equations symbolically by construction and can be trained end-to-end from sparse data within seconds. We prove a universal approximation result showing that this exact model class remains universal on arbitrary domains. FLASH-MAX reaches sub-1% relative validation error from about 1K sparse pointwise observations in seconds, all while maintaining a zero PDE residual, and keeps single-digit errors even for only 100 observations sampled from 3D space. These results suggest that moving governing structure from the loss into the hypothesis class can dramatically improve the trade-off between precision and optimization speed in scientific machine learning.

Alvin Wan, Daniel Ho, Younjin Song et al.

The black-box nature of neural networks limits model decision interpretability, in particular for high-dimensional inputs in computer vision and for dense pixel prediction tasks like segmentation. To address this, prior work combines neural networks with decision trees. However, such models (1) perform poorly when compared to state-of-the-art segmentation models or (2) fail to produce decision rules with spatially-grounded semantic meaning. In this work, we build a hybrid neural-network and decision-tree model for segmentation that (1) attains neural network segmentation accuracy and (2) provides semi-automatically constructed visual decision rules such as "Is there a window?". We obtain semantic visual meaning by extending saliency methods to segmentation and attain accuracy by leveraging insights from neural-backed decision trees, a deep learning analog of decision trees for image classification. Our model SegNBDT attains accuracy within ~2-4% of the state-of-the-art HRNetV2 segmentation model while also retaining explainability; we achieve state-of-the-art performance for explainable models on three benchmark datasets -- Pascal-Context (49.12%), Cityscapes (79.01%), and Look Into Person (51.64%). Furthermore, user studies suggest visual decision rules are more interpretable, particularly for incorrect predictions. Code and pretrained models can be found at https://github.com/daniel-ho/SegNBDT.

Alvin Wan, Lisa Dunlap, Daniel Ho et al.

Machine learning applications such as finance and medicine demand accurate and justifiable predictions, barring most deep learning methods from use. In response, previous work combines decision trees with deep learning, yielding models that (1) sacrifice interpretability for accuracy or (2) sacrifice accuracy for interpretability. We forgo this dilemma by jointly improving accuracy and interpretability using Neural-Backed Decision Trees (NBDTs). NBDTs replace a neural network's final linear layer with a differentiable sequence of decisions and a surrogate loss. This forces the model to learn high-level concepts and lessens reliance on highly-uncertain decisions, yielding (1) accuracy: NBDTs match or outperform modern neural networks on CIFAR, ImageNet and better generalize to unseen classes by up to 16%. Furthermore, our surrogate loss improves the original model's accuracy by up to 2%. NBDTs also afford (2) interpretability: improving human trustby clearly identifying model mistakes and assisting in dataset debugging. Code and pretrained NBDTs are at https://github.com/alvinwan/neural-backed-decision-trees.

Nataniel Ruiz, Sarah Adel Bargal, Stan Sclaroff

Face modification systems using deep learning have become increasingly powerful and accessible. Given images of a person's face, such systems can generate new images of that same person under different expressions and poses. Some systems can also modify targeted attributes such as hair color or age. This type of manipulated images and video have been coined Deepfakes. In order to prevent a malicious user from generating modified images of a person without their consent we tackle the new problem of generating adversarial attacks against such image translation systems, which disrupt the resulting output image. We call this problem disrupting deepfakes. Most image translation architectures are generative models conditioned on an attribute (e.g. put a smile on this person's face). We are first to propose and successfully apply (1) class transferable adversarial attacks that generalize to different classes, which means that the attacker does not need to have knowledge about the conditioning class, and (2) adversarial training for generative adversarial networks (GANs) as a first step towards robust image translation networks. Finally, in gray-box scenarios, blurring can mount a successful defense against disruption. We present a spread-spectrum adversarial attack, which evades blur defenses. Our open-source code can be found at https://github.com/natanielruiz/disrupting-deepfakes.

Hao Yu, Rupayan Mallick, Margrit Betke et al.

Cartoon avatars have been widely used in various applications, including social media, online tutoring, and gaming. However, existing cartoon avatar datasets and generation methods struggle to present highly expressive avatars with fine-grained facial expressions and are often inspired from real-world identities, raising privacy concerns. To address these challenges, we propose a novel framework, GenEAva, for generating high-quality cartoon avatars with fine-grained facial expressions. Our approach fine-tunes a state-of-the-art text-to-image diffusion model to synthesize highly detailed and expressive facial expressions. We then incorporate a stylization model that transforms these realistic faces into cartoon avatars while preserving both identity and expression. Leveraging this framework, we introduce the first expressive cartoon avatar dataset, GenEAva 1.0, specifically designed to capture 135 fine-grained facial expressions, featuring 13,230 expressive cartoon avatars with a balanced distribution across genders, racial groups, and age ranges. We demonstrate that our fine-tuned model generates more expressive faces than the state-of-the-art text-to-image diffusion model SDXL. We also verify that the cartoon avatars generated by our framework do not include memorized identities from fine-tuning data. The proposed framework and dataset provide a diverse and expressive benchmark for future research in cartoon avatar generation.

Hao Yu, Rupayan Mallick, Margrit Betke et al.

Different forms of customized 2D avatars are widely used in gaming applications, virtual communication, education, and content creation. However, existing approaches often fail to capture fine-grained facial expressions and struggle to preserve identity across different expressions. We propose GEN-AFFECT, a novel framework for personalized avatar generation that generates expressive and identity-consistent avatars with a diverse set of facial expressions. Our framework proposes conditioning a multimodal diffusion transformer on an extracted identity-expression representation. This enables identity preservation and representation of a wide range of facial expressions. GEN-AFFECT additionally employs consistent attention at inference for information sharing across the set of generated expressions, enabling the generation process to maintain identity consistency over the array of generated fine-grained expressions. GEN-AFFECT demonstrates superior performance compared to previous state-of-the-art methods on the basis of the accuracy of the generated expressions, the preservation of the identity and the consistency of the target identity across an array of fine-grained facial expressions.

Amro Abdalla, Ismail Shaheen, Dan DeGenaro et al.

We present GIFT: a {G}radient-aware {I}mmunization technique to defend diffusion models against malicious {F}ine-{T}uning while preserving their ability to generate safe content. Existing safety mechanisms like safety checkers are easily bypassed, and concept erasure methods fail under adversarial fine-tuning. GIFT addresses this by framing immunization as a bi-level optimization problem: the upper-level objective degrades the model's ability to represent harmful concepts using representation noising and maximization, while the lower-level objective preserves performance on safe data. GIFT achieves robust resistance to malicious fine-tuning while maintaining safe generative quality. Experimental results show that our method significantly impairs the model's ability to re-learn harmful concepts while maintaining performance on safe content, offering a promising direction for creating inherently safer generative models resistant to adversarial fine-tuning attacks.

Sibo Dong, Ismail Shaheen, Maggie Shen et al.

Text-to-image diffusion models have achieved remarkable success, yet generating coherent image sequences for visual storytelling remains challenging. A key challenge is effectively leveraging all previous text-image pairs, referred to as history text-image pairs, which provide contextual information for maintaining consistency across frames. Existing auto-regressive methods condition on all past image-text pairs but require extensive training, while training-free subject-specific approaches ensure consistency but lack adaptability to narrative prompts. To address these limitations, we propose a multi-modal history adapter for text-to-image diffusion models, \textbf{ViSTA}. It consists of (1) a multi-modal history fusion module to extract relevant history features and (2) a history adapter to condition the generation on the extracted relevant features. We also introduce a salient history selection strategy during inference, where the most salient history text-image pair is selected, improving the quality of the conditioning. Furthermore, we propose to employ a Visual Question Answering-based metric TIFA to assess text-image alignment in visual storytelling, providing a more targeted and interpretable assessment of generated images. Evaluated on the StorySalon and FlintStonesSV dataset, our proposed ViSTA model is not only consistent across different frames, but also well-aligned with the narrative text descriptions.

Klim Kireev, Ana-Maria Creţu, Raphael Meier et al.

Platforms and the law regulate digital content depicting minors (defined as individuals under 18 years of age) differently from other types of content. Given the sheer amount of content that needs to be assessed, machine learning-based automation tools are commonly used to detect content depicting minors. To our knowledge, no dataset or benchmark currently exists for detecting these identification methods in a multi-modal environment. To fill this gap, we release the Image-Caption Children in the Wild Dataset (ICCWD), an image-caption dataset aimed at benchmarking tools that detect depictions of minors. Our dataset is richer than previous child image datasets, containing images of children in a variety of contexts, including fictional depictions and partially visible bodies. ICCWD contains 10,000 image-caption pairs manually labeled to indicate the presence or absence of a child in the image. To demonstrate the possible utility of our dataset, we use it to benchmark three different detectors, including a commercial age estimation system applied to images. Our results suggest that child detection is a challenging task, with the best method achieving a 75.3% true positive rate. We hope the release of our dataset will aid in the design of better minor detection methods in a wide range of scenarios.

Rupayan Mallick, Sibo Dong, Nataniel Ruiz et al.

Applications of diffusion models for visual tasks have been quite noteworthy. This paper targets making classification models more robust to occlusions for the task of object recognition by proposing a pipeline that utilizes a frozen diffusion model. Diffusion features have demonstrated success in image generation and image completion while understanding image context. Occlusion can be posed as an image completion problem by deeming the pixels of the occluder to be `missing.' We hypothesize that such features can help hallucinate object visual features behind occluding objects, and hence we propose using them to enable models to become more occlusion robust. We design experiments to include input-based augmentations as well as feature-based augmentations. Input-based augmentations involve finetuning on images where the occluder pixels are inpainted, and feature-based augmentations involve augmenting classification features with intermediate diffusion features. We demonstrate that our proposed use of diffusion-based features results in models that are more robust to partial object occlusions for both Transformers and ConvNets on ImageNet with simulated occlusions. We also propose a dataset that encompasses real-world occlusions and demonstrate that our method is more robust to partial object occlusions.

Rupayan Mallick, Amr Abdalla, Sarah Adel Bargal

We present FaithFill, a diffusion-based inpainting object completion approach for realistic generation of missing object parts. Typically, multiple reference images are needed to achieve such realistic generation, otherwise the generation would not faithfully preserve shape, texture, color, and background. In this work, we propose a pipeline that utilizes only a single input reference image -having varying lighting, background, object pose, and/or viewpoint. The singular reference image is used to generate multiple views of the object to be inpainted. We demonstrate that FaithFill produces faithful generation of the object's missing parts, together with background/scene preservation, from a single reference image. This is demonstrated through standard similarity metrics, human judgement, and GPT evaluation. Our results are presented on the DreamBooth dataset, and a novel proposed dataset.

Benjamin Spetter-Goldstein, Nataniel Ruiz, Sarah Adel Bargal

The modern open internet contains billions of public images of human faces across the web, especially on social media websites used by half the world's population. In this context, Face Recognition (FR) systems have the potential to match faces to specific names and identities, creating glaring privacy concerns. Adversarial attacks are a promising way to grant users privacy from FR systems by disrupting their capability to recognize faces. Yet, such attacks can be perceptible to human observers, especially under the more challenging black-box threat model. In the literature, the justification for the imperceptibility of such attacks hinges on bounding metrics such as $\ell_p$ norms. However, there is not much research on how these norms match up with human perception. Through examining and measuring both the effectiveness of recent black-box attacks in the face recognition setting and their corresponding human perceptibility through survey data, we demonstrate the trade-offs in perceptibility that occur as attacks become more aggressive. We also show how the $\ell_2$ norm and other metrics do not correlate with human perceptibility in a linear fashion, thus making these norms suboptimal at measuring adversarial attack perceptibility.

Nataniel Ruiz, Adam Kortylewski, Weichao Qiu et al.

Most machine learning models are validated and tested on fixed datasets. This can give an incomplete picture of the capabilities and weaknesses of the model. Such weaknesses can be revealed at test time in the real world. The risks involved in such failures can be loss of profits, loss of time or even loss of life in certain critical applications. In order to alleviate this issue, simulators can be controlled in a fine-grained manner using interpretable parameters to explore the semantic image manifold. In this work, we propose a framework for learning how to test machine learning algorithms using simulators in an adversarial manner in order to find weaknesses in the model before deploying it in critical scenarios. We apply this method in a face recognition setup. We show that certain weaknesses of models trained on real data can be discovered using simulated samples. Using our proposed method, we can find adversarial synthetic faces that fool contemporary face recognition models. This demonstrates the fact that these models have weaknesses that are not measured by commonly used validation datasets. We hypothesize that this type of adversarial examples are not isolated, but usually lie in connected spaces in the latent space of the simulator. We present a method to find these adversarial regions as opposed to the typical adversarial points found in the adversarial example literature.

Dina Bashkirova, Mohamed Abdelfattah, Ziliang Zhu et al.

Less than 35% of recyclable waste is being actually recycled in the US, which leads to increased soil and sea pollution and is one of the major concerns of environmental researchers as well as the common public. At the heart of the problem are the inefficiencies of the waste sorting process (separating paper, plastic, metal, glass, etc.) due to the extremely complex and cluttered nature of the waste stream. Recyclable waste detection poses a unique computer vision challenge as it requires detection of highly deformable and often translucent objects in cluttered scenes without the kind of context information usually present in human-centric datasets. This challenging computer vision task currently lacks suitable datasets or methods in the available literature. In this paper, we take a step towards computer-aided waste detection and present the first in-the-wild industrial-grade waste detection and segmentation dataset, ZeroWaste. We believe that ZeroWaste will catalyze research in object detection and semantic segmentation in extreme clutter as well as applications in the recycling domain. Our project page can be found at http://ai.bu.edu/zerowaste/.

Nataniel Ruiz, Sarah Adel Bargal, Stan Sclaroff

In this work, we develop efficient disruptions of black-box image translation deepfake generation systems. We are the first to demonstrate black-box deepfake generation disruption by presenting image translation formulations of attacks initially proposed for classification models. Nevertheless, a naive adaptation of classification black-box attacks results in a prohibitive number of queries for image translation systems in the real-world. We present a frustratingly simple yet highly effective algorithm Leaking Universal Perturbations (LUP), that significantly reduces the number of queries needed to attack an image. LUP consists of two phases: (1) a short leaking phase where we attack the network using traditional black-box attacks and gather information on successful attacks on a small dataset and (2) and an exploitation phase where we leverage said information to subsequently attack the network with improved efficiency. Our attack reduces the total number of queries necessary to attack GANimation and StarGAN by 30%.

Andrea Zunino, Sarah Adel Bargal, Riccardo Volpi et al.

Conventionally, AI models are thought to trade off explainability for lower accuracy. We develop a training strategy that not only leads to a more explainable AI system for object classification, but as a consequence, suffers no perceptible accuracy degradation. Explanations are defined as regions of visual evidence upon which a deep classification network makes a decision. This is represented in the form of a saliency map conveying how much each pixel contributed to the network's decision. Our training strategy enforces a periodic saliency-based feedback to encourage the model to focus on the image regions that directly correspond to the ground-truth object. We quantify explainability using an automated metric, and using human judgement. We propose explainability as a means for bridging the visual-semantic gap between different domains where model explanations are used as a means of disentagling domain specific information from otherwise relevant features. We demonstrate that this leads to improved generalization to new domains without hindering performance on the original domain.

Nuno C. Garcia, Sarah Adel Bargal, Vitaly Ablavsky et al.

In this work, we address the problem of learning an ensemble of specialist networks using multimodal data, while considering the realistic and challenging scenario of possible missing modalities at test time. Our goal is to leverage the complementary information of multiple modalities to the benefit of the ensemble and each individual network. We introduce a novel Distillation Multiple Choice Learning framework for multimodal data, where different modality networks learn in a cooperative setting from scratch, strengthening one another. The modality networks learned using our method achieve significantly higher accuracy than if trained separately, due to the guidance of other modalities. We evaluate this approach on three video action recognition benchmark datasets. We obtain state-of-the-art results in comparison to other approaches that work with missing modalities at test time.

Donghyun Kim, Sarah Adel Bargal, Jianming Zhang et al.

Deep models are state-of-the-art for many computer vision tasks including image classification and object detection. However, it has been shown that deep models are vulnerable to adversarial examples. We highlight how one-hot encoding directly contributes to this vulnerability and propose breaking away from this widely-used, but highly-vulnerable mapping. We demonstrate that by leveraging a different output encoding, multi-way encoding, we decorrelate source and target models, making target models more secure. Our approach makes it more difficult for adversaries to find useful gradients for generating adversarial attacks. We present robustness for black-box and white-box attacks on four benchmark datasets: MNIST, CIFAR-10, CIFAR-100, and SVHN. The strength of our approach is also presented in the form of an attack for model watermarking, raising challenges in detecting stolen models.

Sarah Adel Bargal, Andrea Zunino, Vitali Petsiuk et al.

We propose Guided Zoom, an approach that utilizes spatial grounding of a model's decision to make more informed predictions. It does so by making sure the model has "the right reasons" for a prediction, defined as reasons that are coherent with those used to make similar correct decisions at training time. The reason/evidence upon which a deep convolutional neural network makes a prediction is defined to be the spatial grounding, in the pixel space, for a specific class conditional probability in the model output. Guided Zoom examines how reasonable such evidence is for each of the top-k predicted classes, rather than solely trusting the top-1 prediction. We show that Guided Zoom improves the classification accuracy of a deep convolutional neural network model and obtains state-of-the-art results on three fine-grained classification benchmark datasets.

Andrea Zunino, Sarah Adel Bargal, Pietro Morerio et al.

We propose a guided dropout regularizer for deep networks based on the evidence of a network prediction defined as the firing of neurons in specific paths. In this work, we utilize the evidence at each neuron to determine the probability of dropout, rather than dropping out neurons uniformly at random as in standard dropout. In essence, we dropout with higher probability those neurons which contribute more to decision making at training time. This approach penalizes high saliency neurons that are most relevant for model prediction, i.e. those having stronger evidence. By dropping such high-saliency neurons, the network is forced to learn alternative paths in order to maintain loss minimization, resulting in a plasticity-like behavior, a characteristic of human brains too. We demonstrate better generalization ability, an increased utilization of network neurons, and a higher resilience to network compression using several metrics over four image/video recognition benchmarks.

Fatih Cakir, Kun He, Sarah Adel Bargal et al.

Binary vector embeddings enable fast nearest neighbor retrieval in large databases of high-dimensional objects, and play an important role in many practical applications, such as image and video retrieval. We study the problem of learning binary vector embeddings under a supervised setting, also known as hashing. We propose a novel supervised hashing method based on optimizing an information-theoretic quantity: mutual information. We show that optimizing mutual information can reduce ambiguity in the induced neighborhood structure in the learned Hamming space, which is essential in obtaining high retrieval performance. To this end, we optimize mutual information in deep neural networks with minibatch stochastic gradient descent, with a formulation that maximally and efficiently utilizes available supervision. Experiments on four image retrieval benchmarks, including ImageNet, confirm the effectiveness of our method in learning high-quality binary embeddings for nearest neighbor retrieval.

Mathew Monfort, Alex Andonian, Bolei Zhou et al.

We present the Moments in Time Dataset, a large-scale human-annotated collection of one million short videos corresponding to dynamic events unfolding within three seconds. Modeling the spatial-audio-temporal dynamics even for actions occurring in 3 second videos poses many challenges: meaningful events do not include only people, but also objects, animals, and natural phenomena; visual and auditory events can be symmetrical in time ("opening" is "closing" in reverse), and either transient or sustained. We describe the annotation process of our dataset (each video is tagged with one action or activity label among 339 different classes), analyze its scale and diversity in comparison to other large-scale video datasets for action recognition, and report results of several baseline models addressing separately, and jointly, three modalities: spatial, temporal and auditory. The Moments in Time dataset, designed to have a large coverage and diversity of events in both visual and auditory modalities, can serve as a new challenge to develop models that scale to the level of complexity and abstract reasoning that a human processes on a daily basis.

Sarah Adel Bargal, Andrea Zunino, Donghyun Kim et al.

Deep models are state-of-the-art for many vision tasks including video action recognition and video captioning. Models are trained to caption or classify activity in videos, but little is known about the evidence used to make such decisions. Grounding decisions made by deep networks has been studied in spatial visual content, giving more insight into model predictions for images. However, such studies are relatively lacking for models of spatiotemporal visual content - videos. In this work, we devise a formulation that simultaneously grounds evidence in space and time, in a single pass, using top-down saliency. We visualize the spatiotemporal cues that contribute to a deep model's classification/captioning output using the model's internal representation. Based on these spatiotemporal cues, we are able to localize segments within a video that correspond with a specific action, or phrase from a caption, without explicitly optimizing/training for these tasks.

Kun He, Fatih Cakir, Sarah Adel Bargal et al.

Hashing, or learning binary embeddings of data, is frequently used in nearest neighbor retrieval. In this paper, we develop learning to rank formulations for hashing, aimed at directly optimizing ranking-based evaluation metrics such as Average Precision (AP) and Normalized Discounted Cumulative Gain (NDCG). We first observe that the integer-valued Hamming distance often leads to tied rankings, and propose to use tie-aware versions of AP and NDCG to evaluate hashing for retrieval. Then, to optimize tie-aware ranking metrics, we derive their continuous relaxations, and perform gradient-based optimization with deep neural networks. Our results establish the new state-of-the-art for image retrieval by Hamming ranking in common benchmarks.

Fatih Cakir, Kun He, Sarah Adel Bargal et al.

Learning-based hashing methods are widely used for nearest neighbor retrieval, and recently, online hashing methods have demonstrated good performance-complexity trade-offs by learning hash functions from streaming data. In this paper, we first address a key challenge for online hashing: the binary codes for indexed data must be recomputed to keep pace with updates to the hash functions. We propose an efficient quality measure for hash functions, based on an information-theoretic quantity, mutual information, and use it successfully as a criterion to eliminate unnecessary hash table updates. Next, we also show how to optimize the mutual information objective using stochastic gradient descent. We thus develop a novel hashing method, MIHash, that can be used in both online and batch settings. Experiments on image retrieval benchmarks (including a 2.5M image dataset) confirm the effectiveness of our formulation, both in reducing hash table recomputations and in learning high-quality hash functions.

Shugao Ma, Sarah Adel Bargal, Jianming Zhang et al.

Recently, attempts have been made to collect millions of videos to train CNN models for action recognition in videos. However, curating such large-scale video datasets requires immense human labor, and training CNNs on millions of videos demands huge computational resources. In contrast, collecting action images from the Web is much easier and training on images requires much less computation. In addition, labeled web images tend to contain discriminative action poses, which highlight discriminative portions of a video's temporal progression. We explore the question of whether we can utilize web action images to train better CNN models for action recognition in videos. We collect 23.8K manually filtered images from the Web that depict the 101 actions in the UCF101 action video dataset. We show that by utilizing web action images along with videos in training, significant performance boosts of CNN models can be achieved. We then investigate the scalability of the process by leveraging crawled web images (unfiltered) for UCF101 and ActivityNet. We replace 16.2M video frames by 393K unfiltered images and get comparable performance.

Fatih Cakir, Sarah Adel Bargal, Stan Sclaroff

Supervised hashing methods are widely-used for nearest neighbor search in computer vision applications. Most state-of-the-art supervised hashing approaches employ batch-learners. Unfortunately, batch-learning strategies can be inefficient when confronted with large training datasets. Moreover, with batch-learners, it is unclear how to adapt the hash functions as a dataset continues to grow and diversify over time. Yet, in many practical scenarios the dataset grows and diversifies; thus, both the hash functions and the indexing must swiftly accommodate these changes. To address these issues, we propose an online hashing method that is amenable to changes and expansions of the datasets. Since it is an online algorithm, our approach offers linear complexity with the dataset size. Our solution is supervised, in that we incorporate available label information to preserve the semantic neighborhood. Such an adaptive hashing method is attractive; but it requires recomputing the hash table as the hash functions are updated. If the frequency of update is high, then recomputing the hash table entries may cause inefficiencies in the system, especially for large indexes. Thus, we also propose a framework to reduce hash table updates. We compare our method to state-of-the-art solutions on two benchmarks and demonstrate significant improvements over previous work.