Kate Saenko

Nicolas Carion, Laura Gustafson, Yuan-Ting Hu et al.

We present Segment Anything Model (SAM) 3, a unified model that detects, segments, and tracks objects in images and videos based on concept prompts, which we define as either short noun phrases (e.g., "yellow school bus"), image exemplars, or a combination of both. Promptable Concept Segmentation (PCS) takes such prompts and returns segmentation masks and unique identities for all matching object instances. To advance PCS, we build a scalable data engine that produces a high-quality dataset with 4M unique concept labels, including hard negatives, across images and videos. Our model consists of an image-level detector and a memory-based video tracker that share a single backbone. Recognition and localization are decoupled with a presence head, which boosts detection accuracy. SAM 3 doubles the accuracy of existing systems in both image and video PCS, and improves previous SAM capabilities on visual segmentation tasks. We open source SAM 3 along with our new Segment Anything with Concepts (SA-Co) benchmark for promptable concept segmentation.

Kuniaki Saito, Kihyuk Sohn, Xiang Zhang et al.

In Composed Image Retrieval (CIR), a user combines a query image with text to describe their intended target. Existing methods rely on supervised learning of CIR models using labeled triplets consisting of the query image, text specification, and the target image. Labeling such triplets is expensive and hinders broad applicability of CIR. In this work, we propose to study an important task, Zero-Shot Composed Image Retrieval (ZS-CIR), whose goal is to build a CIR model without requiring labeled triplets for training. To this end, we propose a novel method, called Pic2Word, that requires only weakly labeled image-caption pairs and unlabeled image datasets to train. Unlike existing supervised CIR models, our model trained on weakly labeled or unlabeled datasets shows strong generalization across diverse ZS-CIR tasks, e.g., attribute editing, object composition, and domain conversion. Our approach outperforms several supervised CIR methods on the common CIR benchmark, CIRR and Fashion-IQ. Code will be made publicly available at https://github.com/google-research/composed_image_retrieval.

Donghyun Kim, Kaihong Wang, Kate Saenko et al.

While pose estimation is an important computer vision task, it requires expensive annotation and suffers from domain shift. In this paper, we investigate the problem of domain adaptive 2D pose estimation that transfers knowledge learned on a synthetic source domain to a target domain without supervision. While several domain adaptive pose estimation models have been proposed recently, they are not generic but only focus on either human pose or animal pose estimation, and thus their effectiveness is somewhat limited to specific scenarios. In this work, we propose a unified framework that generalizes well on various domain adaptive pose estimation problems. We propose to align representations using both input-level and output-level cues (pixels and pose labels, respectively), which facilitates the knowledge transfer from the source domain to the unlabeled target domain. Our experiments show that our method achieves state-of-the-art performance under various domain shifts. Our method outperforms existing baselines on human pose estimation by up to 4.5 percent points (pp), hand pose estimation by up to 7.4 pp, and animal pose estimation by up to 4.8 pp for dogs and 3.3 pp for sheep. These results suggest that our method is able to mitigate domain shift on diverse tasks and even unseen domains and objects (e.g., trained on horse and tested on dog). Our code will be publicly available at: https://github.com/VisionLearningGroup/UDA_PoseEstimation.

Maan Qraitem, Kate Saenko, Bryan A. Plummer

Prior work has shown that Visual Recognition datasets frequently underrepresent bias groups $B$ (\eg Female) within class labels $Y$ (\eg Programmers). This dataset bias can lead to models that learn spurious correlations between class labels and bias groups such as age, gender, or race. Most recent methods that address this problem require significant architectural changes or additional loss functions requiring more hyper-parameter tuning. Alternatively, data sampling baselines from the class imbalance literature (\eg Undersampling, Upweighting), which can often be implemented in a single line of code and often have no hyperparameters, offer a cheaper and more efficient solution. However, these methods suffer from significant shortcomings. For example, Undersampling drops a significant part of the input distribution per epoch while Oversampling repeats samples, causing overfitting. To address these shortcomings, we introduce a new class-conditioned sampling method: Bias Mimicking. The method is based on the observation that if a class $c$ bias distribution, \ie $P_D(B|Y=c)$ is mimicked across every $c^{\prime}\neq c$, then $Y$ and $B$ are statistically independent. Using this notion, BM, through a novel training procedure, ensures that the model is exposed to the entire distribution per epoch without repeating samples. Consequently, Bias Mimicking improves underrepresented groups' accuracy of sampling methods by 3\% over four benchmarks while maintaining and sometimes improving performance over nonsampling methods. Code: \url{https://github.com/mqraitem/Bias-Mimicking}

Piotr Teterwak, Kuniaki Saito, Theodoros Tsiligkaridis et al.

Domain Generalization (DG) aims to develop classifiers that can generalize to new, unseen data distributions, a critical capability when collecting new domain-specific data is impractical. A common DG baseline minimizes the empirical risk on the source domains. Recent studies have shown that this approach, known as Empirical Risk Minimization (ERM), can outperform most more complex DG methods when properly tuned. However, these studies have primarily focused on a narrow set of hyperparameters, neglecting other factors that can enhance robustness and prevent overfitting and catastrophic forgetting, properties which are critical for strong DG performance. In our investigation of training data utilization (i.e., duration and setting validation splits), initialization, and additional regularizers, we find that tuning these previously overlooked factors significantly improves model generalization across diverse datasets without adding much complexity. We call this improved, yet simple baseline ERM++. Despite its ease of implementation, ERM++ improves DG performance by over 5\% compared to prior ERM baselines on a standard benchmark of 5 datasets with a ResNet-50 and over 15\% with a ViT-B/16. It also outperforms all state-of-the-art methods on DomainBed datasets with both architectures. Importantly, ERM++ is easy to integrate into existing frameworks like DomainBed, making it a practical and powerful tool for researchers and practitioners. Overall, ERM++ challenges the need for more complex DG methods by providing a stronger, more reliable baseline that maintains simplicity and ease of use. Code is available at \url{https://github.com/piotr-teterwak/erm_plusplus}

Maan Qraitem, Kate Saenko, Bryan A. Plummer

Visual recognition models are prone to learning spurious correlations induced by a biased training set where certain conditions $B$ (\eg, Indoors) are over-represented in certain classes $Y$ (\eg, Big Dogs). Synthetic data from off-the-shelf large-scale generative models offers a promising direction to mitigate this issue by augmenting underrepresented subgroups in the real dataset. However, by using a mixed distribution of real and synthetic data, we introduce another source of bias due to distributional differences between synthetic and real data (\eg synthetic artifacts). As we will show, prior work's approach for using synthetic data to resolve the model's bias toward $B$ do not correct the model's bias toward the pair $(B, G)$, where $G$ denotes whether the sample is real or synthetic. Thus, the model could simply learn signals based on the pair $(B, G)$ (\eg, Synthetic Indoors) to make predictions about $Y$ (\eg, Big Dogs). To address this issue, we propose a simple, easy-to-implement, two-step training pipeline that we call From Fake to Real (FFR). The first step of FFR pre-trains a model on balanced synthetic data to learn robust representations across subgroups. In the second step, FFR fine-tunes the model on real data using ERM or common loss-based bias mitigation methods. By training on real and synthetic data separately, FFR does not expose the model to the statistical differences between real and synthetic data and thus avoids the issue of bias toward the pair $(B, G)$. Our experiments show that FFR improves worst group accuracy over the state-of-the-art by up to 20\% over three datasets. Code available: \url{https://github.com/mqraitem/From-Fake-to-Real}

Kuniaki Saito, Donghyun Kim, Piotr Teterwak et al.

Building object detectors that are robust to domain shifts is critical for real-world applications. Prior approaches fine-tune a pre-trained backbone and risk overfitting it to in-distribution (ID) data and distorting features useful for out-of-distribution (OOD) generalization. We propose to use Relative Gradient Norm (RGN) as a way to measure the vulnerability of a backbone to feature distortion, and show that high RGN is indeed correlated with lower OOD performance. Our analysis of RGN yields interesting findings: some backbones lose OOD robustness during fine-tuning, but others gain robustness because their architecture prevents the parameters from changing too much from the initial model. Given these findings, we present recipes to boost OOD robustness for both types of backbones. Specifically, we investigate regularization and architectural choices for minimizing gradient updates so as to prevent the tuned backbone from losing generalizable features. Our proposed techniques complement each other and show substantial improvements over baselines on diverse architectures and datasets. Code is available at https://github.com/VisionLearningGroup/mind_back.

Ximeng Sun, Ping Hu, Kate Saenko

Solving multi-label recognition (MLR) for images in the low-label regime is a challenging task with many real-world applications. Recent work learns an alignment between textual and visual spaces to compensate for insufficient image labels, but loses accuracy because of the limited amount of available MLR annotations. In this work, we utilize the strong alignment of textual and visual features pretrained with millions of auxiliary image-text pairs and propose Dual Context Optimization (DualCoOp) as a unified framework for partial-label MLR and zero-shot MLR. DualCoOp encodes positive and negative contexts with class names as part of the linguistic input (i.e. prompts). Since DualCoOp only introduces a very light learnable overhead upon the pretrained vision-language framework, it can quickly adapt to multi-label recognition tasks that have limited annotations and even unseen classes. Experiments on standard multi-label recognition benchmarks across two challenging low-label settings demonstrate the advantages of our approach over state-of-the-art methods.

Donghyun Kim, Kaihong Wang, Stan Sclaroff et al.

Deep models must learn robust and transferable representations in order to perform well on new domains. While domain transfer methods (e.g., domain adaptation, domain generalization) have been proposed to learn transferable representations across domains, they are typically applied to ResNet backbones pre-trained on ImageNet. Thus, existing works pay little attention to the effects of pre-training on domain transfer tasks. In this paper, we provide a broad study and in-depth analysis of pre-training for domain adaptation and generalization, namely: network architectures, size, pre-training loss, and datasets. We observe that simply using a state-of-the-art backbone outperforms existing state-of-the-art domain adaptation baselines and set new baselines on Office-Home and DomainNet improving by 10.7\% and 5.5\%. We hope that this work can provide more insights for future domain transfer research.

Ping Hu, Simon Niklaus, Stan Sclaroff et al.

Motion-based video frame interpolation commonly relies on optical flow to warp pixels from the inputs to the desired interpolation instant. Yet due to the inherent challenges of motion estimation (e.g. occlusions and discontinuities), most state-of-the-art interpolation approaches require subsequent refinement of the warped result to generate satisfying outputs, which drastically decreases the efficiency for multi-frame interpolation. In this work, we propose a fully differentiable Many-to-Many (M2M) splatting framework to interpolate frames efficiently. Specifically, given a frame pair, we estimate multiple bidirectional flows to directly forward warp the pixels to the desired time step, and then fuse any overlapping pixels. In doing so, each source pixel renders multiple target pixels and each target pixel can be synthesized from a larger area of visual context. This establishes a many-to-many splatting scheme with robustness to artifacts like holes. Moreover, for each input frame pair, M2M only performs motion estimation once and has a minuscule computational overhead when interpolating an arbitrary number of in-between frames, hence achieving fast multi-frame interpolation. We conducted extensive experiments to analyze M2M, and found that it significantly improves efficiency while maintaining high effectiveness.

Eunice Yiu, Maan Qraitem, Anisa Noor Majhi et al. · berkeley

This paper investigates visual analogical reasoning in large multimodal models (LMMs) compared to human adults and children. A "visual analogy" is an abstract rule inferred from one image and applied to another. While benchmarks exist for testing visual reasoning in LMMs, they require advanced skills and omit basic visual analogies that even young children can make. Inspired by developmental psychology, we propose a new benchmark of 4,300 visual transformations of everyday objects to test LMMs on visual analogical reasoning and compare them to children (ages three to five) and to adults. We structure the evaluation into three stages: identifying what changed (e.g., color, number, etc.), how it changed (e.g., added one object), and applying the rule to new scenarios. Our findings show that while GPT-o1, GPT-4V, LLaVA-1.5, and MANTIS identify the "what" effectively, they struggle with quantifying the "how" and extrapolating this rule to new objects. In contrast, children and adults exhibit much stronger analogical reasoning at all three stages. Additionally, the strongest tested model, GPT-o1, performs better in tasks involving simple surface-level visual attributes like color and size, correlating with quicker human adult response times. Conversely, more complex tasks such as number, rotation, and reflection, which necessitate extensive cognitive processing and understanding of extrinsic spatial properties in the physical world, present more significant challenges. Altogether, these findings highlight the limitations of training models on data that primarily consists of 2D images and text.

Dina Bashkirova, Jose Lezama, Kihyuk Sohn et al.

Recent conditional image generation methods produce images of remarkable diversity, fidelity and realism. However, the majority of these methods allow conditioning only on labels or text prompts, which limits their level of control over the generation result. In this paper, we introduce MaskSketch, an image generation method that allows spatial conditioning of the generation result using a guiding sketch as an extra conditioning signal during sampling. MaskSketch utilizes a pre-trained masked generative transformer, requiring no model training or paired supervision, and works with input sketches of different levels of abstraction. We show that intermediate self-attention maps of a masked generative transformer encode important structural information of the input image, such as scene layout and object shape, and we propose a novel sampling method based on this observation to enable structure-guided generation. Our results show that MaskSketch achieves high image realism and fidelity to the guiding structure. Evaluated on standard benchmark datasets, MaskSketch outperforms state-of-the-art methods for sketch-to-image translation, as well as unpaired image-to-image translation approaches.

Ximeng Sun, Pengchuan Zhang, Peizhao Zhang et al.

Large Vision-Language Foundation Models (VLFM), such as CLIP, ALIGN and Florence, are trained on large-scale datasets of image-caption pairs and achieve superior transferability and robustness on downstream tasks, but they are difficult to use in many practical applications due to their large size, high latency and fixed architectures. Unfortunately, recent work shows training a small custom VLFM for resource-limited applications is currently very difficult using public and smaller-scale data. In this paper, we introduce a new distillation mechanism (DIME-FM) that allows us to transfer the knowledge contained in large VLFMs to smaller, customized foundation models using a relatively small amount of inexpensive, unpaired images and sentences. We transfer the knowledge from the pre-trained CLIP-ViTL/14 model to a ViT-B/32 model, with only 40M public images and 28.4M unpaired public sentences. The resulting model "Distill-ViT-B/32" rivals the CLIP-ViT-B/32 model pre-trained on its private WiT dataset (400M image-text pairs): Distill-ViT-B/32 achieves similar results in terms of zero-shot and linear-probing performance on both ImageNet and the ELEVATER (20 image classification tasks) benchmarks. It also displays comparable robustness when evaluated on five datasets with natural distribution shifts from ImageNet.

Maan Qraitem, Kate Saenko, Bryan A. Plummer

Procedural content generation has enabled vast virtual worlds through levels, maps, and quests, but large-scale character generation remains underexplored. We identify two alignment-induced biases in existing methods: a positive moral bias, where characters uniformly adopt agreeable stances (e.g. always saying lying is bad), and a helpful assistant bias, where characters invariably answer questions directly (e.g. never refusing or deflecting). While such tendencies suit instruction-following systems, they suppress dramatic tension and yield predictable characters, stemming from maximum likelihood training and assistant fine-tuning. To address this, we introduce PersonaWeaver, a framework that disentangles world-building (roles, demographics) from behavioral-building (moral stances, interactional styles), yielding characters with more diverse reactions and moral stances, as well as second-order diversity in stylistic markers like length, tone, and punctuation. Code: https://github.com/mqraitem/Persona-Weaver

Ping Hu, Ximeng Sun, Stan Sclaroff et al.

Multi-label image recognition in the low-label regime is a task of great challenge and practical significance. Previous works have focused on learning the alignment between textual and visual spaces to compensate for limited image labels, yet may suffer from reduced accuracy due to the scarcity of high-quality multi-label annotations. In this research, we leverage the powerful alignment between textual and visual features pretrained with millions of auxiliary image-text pairs. We introduce an efficient and effective framework called Evidence-guided Dual Context Optimization (DualCoOp++), which serves as a unified approach for addressing partial-label and zero-shot multi-label recognition. In DualCoOp++ we separately encode evidential, positive, and negative contexts for target classes as parametric components of the linguistic input (i.e., prompts). The evidential context aims to discover all the related visual content for the target class, and serves as guidance to aggregate positive and negative contexts from the spatial domain of the image, enabling better distinguishment between similar categories. Additionally, we introduce a Winner-Take-All module that promotes inter-class interaction during training, while avoiding the need for extra parameters and costs. As DualCoOp++ imposes minimal additional learnable overhead on the pretrained vision-language framework, it enables rapid adaptation to multi-label recognition tasks with limited annotations and even unseen classes. Experiments on standard multi-label recognition benchmarks across two challenging low-label settings demonstrate the superior performance of our approach compared to state-of-the-art methods.

Reuben Tan, Arijit Ray, Andrea Burns et al.

We propose a self-supervised approach for learning to perform audio source separation in videos based on natural language queries, using only unlabeled video and audio pairs as training data. A key challenge in this task is learning to associate the linguistic description of a sound-emitting object to its visual features and the corresponding components of the audio waveform, all without access to annotations during training. To overcome this challenge, we adapt off-the-shelf vision-language foundation models to provide pseudo-target supervision via two novel loss functions and encourage a stronger alignment between the audio, visual and natural language modalities. During inference, our approach can separate sounds given text, video and audio input, or given text and audio input alone. We demonstrate the effectiveness of our self-supervised approach on three audio-visual separation datasets, including MUSIC, SOLOS and AudioSet, where we outperform state-of-the-art strongly supervised approaches despite not using object detectors or text labels during training.

Amit Alfassy, Assaf Arbelle, Oshri Halimi et al.

Foundation Models (FMs) have demonstrated unprecedented capabilities including zero-shot learning, high fidelity data synthesis, and out of domain generalization. However, as we show in this paper, FMs still have poor out-of-the-box performance on expert tasks (e.g. retrieval of car manuals technical illustrations from language queries), data for which is either unseen or belonging to a long-tail part of the data distribution of the huge datasets used for FM pre-training. This underlines the necessity to explicitly evaluate and finetune FMs on such expert tasks, arguably ones that appear the most in practical real-world applications. In this paper, we propose a first of its kind FETA benchmark built around the task of teaching FMs to understand technical documentation, via learning to match their graphical illustrations to corresponding language descriptions. Our FETA benchmark focuses on text-to-image and image-to-text retrieval in public car manuals and sales catalogue brochures. FETA is equipped with a procedure for completely automatic annotation extraction (code would be released upon acceptance), allowing easy extension of FETA to more documentation types and application domains in the future. Our automatic annotation leads to an automated performance metric shown to be consistent with metrics computed on human-curated annotations (also released). We provide multiple baselines and analysis of popular FMs on FETA leading to several interesting findings that we believe would be very valuable to the FM community, paving the way towards real-world application of FMs for practical expert tasks currently 'overlooked' by standard benchmarks focusing on common objects.

Kuniaki Saito, Kihyuk Sohn, Xiang Zhang et al.

Image-classification datasets have been used to pretrain image recognition models. Recently, web-scale image-caption datasets have emerged as a source of powerful pretraining alternative. Image-caption datasets are more ``open-domain'', containing a wider variety of scene types and vocabulary words than traditional classification datasets, and models trained on these datasets have demonstrated strong performance on few- and zero-shot recognition tasks. When naively unifying image-classification and -caption dataset, we show that such dataset biases negatively affect pre-training by reducing the generalizability of learned representations and thus jeopardizing zero-shot performance since the unification can tailor the model for the classification dataset, making it vulnerable to the distribution shift from the dataset. In this work, we address the problem by disentangling the dataset bias using prefix tokens that inform a language encoder of the type of the input dataset (e.g., image-classification or caption) at training time. This approach allows the language encoder to share the knowledge from two datasets as well as switch the mode of feature extraction, i.e., image-classification dataset or image-caption dataset tailored mode, where we use image-caption mode in the zero-shot evaluation. Our method is generic and can be easily integrated into existing VL pre-training objectives such as CLIP or UniCL. In experiments, we show that this simple technique improves the performance in zero-shot image recognition accuracy and robustness to the image-level distribution shift.

Vitali Petsiuk, Alexander E. Siemenn, Saisamrit Surbehera et al.

We provide a new multi-task benchmark for evaluating text-to-image models. We perform a human evaluation comparing the most common open-source (Stable Diffusion) and commercial (DALL-E 2) models. Twenty computer science AI graduate students evaluated the two models, on three tasks, at three difficulty levels, across ten prompts each, providing 3,600 ratings. Text-to-image generation has seen rapid progress to the point that many recent models have demonstrated their ability to create realistic high-resolution images for various prompts. However, current text-to-image methods and the broader body of research in vision-language understanding still struggle with intricate text prompts that contain many objects with multiple attributes and relationships. We introduce a new text-to-image benchmark that contains a suite of thirty-two tasks over multiple applications that capture a model's ability to handle different features of a text prompt. For example, asking a model to generate a varying number of the same object to measure its ability to count or providing a text prompt with several objects that each have a different attribute to identify its ability to match objects and attributes correctly. Rather than subjectively evaluating text-to-image results on a set of prompts, our new multi-task benchmark consists of challenge tasks at three difficulty levels (easy, medium, and hard) and human ratings for each generated image.

Katherine Deng, Arijit Ray, Reuben Tan et al.

Existing emotion prediction benchmarks contain coarse emotion labels which do not consider the diversity of emotions that an image and text can elicit in humans due to various reasons. Learning diverse reactions to multimodal content is important as intelligent machines take a central role in generating and delivering content to society. To address this gap, we propose Socratis, a societal reactions benchmark, where each image-caption (IC) pair is annotated with multiple emotions and the reasons for feeling them. Socratis contains 18K free-form reactions for 980 emotions on 2075 image-caption pairs from 5 widely-read news and image-caption (IC) datasets. We benchmark the capability of state-of-the-art multimodal large language models to generate the reasons for feeling an emotion given an IC pair. Based on a preliminary human study, we observe that humans prefer human-written reasons over 2 times more often than machine-generated ones. This shows our task is harder than standard generation tasks because it starkly contrasts recent findings where humans cannot tell apart machine vs human-written news articles, for instance. We further see that current captioning metrics based on large vision-language models also fail to correlate with human preferences. We hope that these findings and our benchmark will inspire further research on training emotionally aware models.

Reuben Tan, Matthias De Lange, Michael Iuzzolino et al.

Long-term activity forecasting is an especially challenging research problem because it requires understanding the temporal relationships between observed actions, as well as the variability and complexity of human activities. Despite relying on strong supervision via expensive human annotations, state-of-the-art forecasting approaches often generalize poorly to unseen data. To alleviate this issue, we propose Multiscale Video Pretraining (MVP), a novel self-supervised pretraining approach that learns robust representations for forecasting by learning to predict contextualized representations of future video clips over multiple timescales. MVP is based on our observation that actions in videos have a multiscale nature, where atomic actions typically occur at a short timescale and more complex actions may span longer timescales. We compare MVP to state-of-the-art self-supervised video learning approaches on downstream long-term forecasting tasks including long-term action anticipation and video summary prediction. Our comprehensive experiments across the Ego4D and Epic-Kitchens-55/100 datasets demonstrate that MVP out-performs state-of-the-art methods by significant margins. Notably, MVP obtains a relative performance gain of over 20% accuracy in video summary forecasting over existing methods.

Ping Hu, Simon Niklaus, Lu Zhang et al.

In this work, we first propose a fully differentiable Many-to-Many (M2M) splatting framework to interpolate frames efficiently. Given a frame pair, we estimate multiple bidirectional flows to directly forward warp the pixels to the desired time step before fusing overlapping pixels. In doing so, each source pixel renders multiple target pixels and each target pixel can be synthesized from a larger area of visual context, establishing a many-to-many splatting scheme with robustness to undesirable artifacts. For each input frame pair, M2M has a minuscule computational overhead when interpolating an arbitrary number of in-between frames, hence achieving fast multi-frame interpolation. However, directly warping and fusing pixels in the intensity domain is sensitive to the quality of motion estimation and may suffer from less effective representation capacity. To improve interpolation accuracy, we further extend an M2M++ framework by introducing a flexible Spatial Selective Refinement (SSR) component, which allows for trading computational efficiency for interpolation quality and vice versa. Instead of refining the entire interpolated frame, SSR only processes difficult regions selected under the guidance of an estimated error map, thereby avoiding redundant computation. Evaluation on multiple benchmark datasets shows that our method is able to improve the efficiency while maintaining competitive video interpolation quality, and it can be adjusted to use more or less compute as needed.

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

Reuben Tan, Bryan A. Plummer, Kate Saenko et al.

Recent self-supervised approaches have used large-scale image-text datasets to learn powerful representations that transfer to many tasks without finetuning. These methods often assume that there is one-to-one correspondence between its images and their (short) captions. However, many tasks require reasoning about multiple images and long text narratives, such as describing news articles with visual summaries. Thus, we explore a novel setting where the goal is to learn a self-supervised visual-language representation that is robust to varying text length and the number of images. In addition, unlike prior work which assumed captions have a literal relation to the image, we assume images only contain loose illustrative correspondence with the text. To explore this problem, we introduce a large-scale multimodal dataset containing over 31M articles, 22M images and 1M videos. We show that state-of-the-art image-text alignment methods are not robust to longer narratives with multiple images. Finally, we introduce an intuitive baseline that outperforms these methods on zero-shot image-set retrieval by 10% on the GoodNews dataset.

Kaihong Wang, Donghyun Kim, Rogerio Feris et al.

While transformers have greatly boosted performance in semantic segmentation, domain adaptive transformers are not yet well explored. We identify that the domain gap can cause discrepancies in self-attention. Due to this gap, the transformer attends to spurious regions or pixels, which deteriorates accuracy on the target domain. We propose to perform adaptation on attention maps with cross-domain attention layers that share features between the source and the target domains. Specifically, we impose consistency between predictions from cross-domain attention and self-attention modules to encourage similar distribution in the attention and output of the model across domains, i.e., attention-level and output-level alignment. We also enforce consistency in attention maps between different augmented views to further strengthen the attention-based alignment. Combining these two components, our method mitigates the discrepancy in attention maps across domains and further boosts the performance of the transformer under unsupervised domain adaptation settings. Our model outperforms the existing state-of-the-art baseline model on three widely used benchmarks, including GTAV-to-Cityscapes by 1.3 percent point (pp), Synthia-to-Cityscapes by 0.6 pp, and Cityscapes-to-ACDC by 1.1 pp, on average. Additionally, we verify the effectiveness and generalizability of our method through extensive experiments. Our code will be publicly available.

Bassel El Mabsout, Shahin Roozkhosh, Siddharth Mysore et al.

Fine-tuning simulation-trained RL agents with real-world data often degrades crucial behaviors due to limited or skewed data distributions. We argue that designer priorities exist not just in reward functions, but also in simulation design choices like task selection and state initialization. When adapting to real-world data, agents can experience catastrophic forgetting in important but underrepresented scenarios. We propose framing live-adaptation as a multi-objective optimization problem, where policy objectives must be satisfied both in simulation and reality. Our approach leverages critics from simulation as "anchors for design intent" (anchor critics). By jointly optimizing policies against both anchor critics and critics trained on real-world experience, our method enables adaptation while preserving prioritized behaviors from simulation. Evaluations demonstrate robust behavior retention in sim-to-sim benchmarks and a sim-to-real scenario with a racing quadrotor, allowing for power consumption reductions of up to 50% without control loss. We also contribute SwaNNFlight, an open-source firmware for enabling live adaptation on similar robotic platforms.

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/

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.

Ximeng Sun, Kihyuk Sohn, Kate Saenko et al.

The cost of labeling data often limits the performance of machine learning systems. In multi-task learning, related tasks provide information to each other and improve overall performance, but the label cost can vary among tasks. How should the label budget (i.e. the amount of money spent on labeling) be allocated among different tasks to achieve optimal multi-task performance? We are the first to propose and formally define the label budget allocation problem in multi-task learning and to empirically show that different budget allocation strategies make a big difference to its performance. We propose a Task-Adaptive Budget Allocation algorithm to robustly generate the optimal budget allocation adaptive to different multi-task learning settings. Specifically, we estimate and then maximize the extent of new information obtained from the allocated budget as a proxy for multi-task learning performance. Experiments on PASCAL VOC and Taskonomy demonstrate the efficacy of our approach over other widely used heuristic labeling strategies.

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.

Arijit Ray, Ahmed Abdelkader, Chengzhi Mao et al.

Reasoning goes beyond language; the real world requires reasoning about space, time, affordances, and much more that words alone cannot convey. Existing multimodal models exploring the potential of reasoning with images are brittle and do not scale. They rely on calling specialist tools, costly generation of images, or handcrafted reasoning data to switch between text and image thoughts. Instead, we offer a simpler alternative -- Mull-Tokens -- modality-agnostic latent tokens pre-trained to hold intermediate information in either image or text modalities to let the model think free-form towards the correct answer. We investigate best practices to train Mull-Tokens inspired by latent reasoning frameworks. We first train Mull-Tokens using supervision from interleaved text-image traces, and then fine-tune without any supervision by only using the final answers. Across four challenging spatial reasoning benchmarks involving tasks such as solving puzzles and taking different perspectives, we demonstrate that Mull-Tokens improve upon several baselines utilizing text-only reasoning or interleaved image-text reasoning, achieving a +3% average improvement and up to +16% on a puzzle solving reasoning-heavy split compared to our strongest baseline. Adding to conversations around challenges in grounding textual and visual reasoning, Mull-Tokens offers a simple solution to abstractly think in multiple modalities.

Shengao Wang, Wenqi Wang, Zecheng Wang et al.

Early children's developmental trajectories set up a natural goal for sample-efficient pretraining of vision foundation models. We introduce BabyVLM-V2, a developmentally grounded framework for infant-inspired vision-language modeling that extensively improves upon BabyVLM-V1 through a longitudinal, multifaceted pretraining set, a versatile model, and, most importantly, DevCV Toolbox for cognitive evaluation. The pretraining set maximizes coverage while minimizing curation of a longitudinal, infant-centric audiovisual corpus, yielding video-utterance, image-utterance, and multi-turn conversational data that mirror infant experiences. DevCV Toolbox adapts all vision-related measures of the recently released NIH Baby Toolbox into a benchmark suite of ten multimodal tasks, covering spatial reasoning, memory, and vocabulary understanding aligned with early children's capabilities. Experimental results show that a compact model pretrained from scratch can achieve competitive performance on DevCV Toolbox, outperforming GPT-4o on some tasks. We hope the principled, unified BabyVLM-V2 framework will accelerate research in developmentally plausible pretraining of vision foundation models.

Maan Qraitem, Nazia Tasnim, Piotr Teterwak et al.

Typographic attacks, adding misleading text to images, can deceive vision-language models (LVLMs). The susceptibility of recent large LVLMs like GPT4-V to such attacks is understudied, raising concerns about amplified misinformation in personal assistant applications. Previous attacks use simple strategies, such as random misleading words, which don't fully exploit LVLMs' language reasoning abilities. We introduce an experimental setup for testing typographic attacks on LVLMs and propose two novel self-generated attacks: (1) Class-based attacks, where the model identifies a similar class to deceive itself, and (2) Reasoned attacks, where an advanced LVLM suggests an attack combining a deceiving class and description. Our experiments show these attacks significantly reduce classification performance by up to 60\% and are effective across different models, including InstructBLIP and MiniGPT4. Code: https://github.com/mqraitem/Self-Gen-Typo-Attack

Samarth Mishra, Carlos D. Castillo, Hongcheng Wang et al.

In cross-domain retrieval, a model is required to identify images from the same semantic category across two visual domains. For instance, given a sketch of an object, a model needs to retrieve a real image of it from an online store's catalog. A standard approach for such a problem is learning a feature space of images where Euclidean distances reflect similarity. Even without human annotations, which may be expensive to acquire, prior methods function reasonably well using unlabeled images for training. Our problem constraint takes this further to scenarios where the two domains do not necessarily share any common categories in training data. This can occur when the two domains in question come from different versions of some biometric sensor recording identities of different people. We posit a simple solution, which is to generate synthetic data to fill in these missing category examples across domains. This, we do via category preserving translation of images from one visual domain to another. We compare approaches specifically trained for this translation for a pair of domains, as well as those that can use large-scale pre-trained text-to-image diffusion models via prompts, and find that the latter can generate better replacement synthetic data, leading to more accurate cross-domain retrieval models. Our best SynCDR model can outperform prior art by up to 15\%. Code for our work is available at https://github.com/samarth4149/SynCDR .

Maan Qraitem, Piotr Teterwak, Kate Saenko et al.

Vision-language models (VLMs) (e.g. CLIP, LLaVA) are trained on large-scale, lightly curated web datasets, leading them to learn unintended correlations between semantic concepts and unrelated visual signals. These associations degrade model accuracy by causing predictions to rely on incidental patterns rather than genuine visual understanding. Prior work has weaponized these correlations as an attack vector to manipulate model predictions, such as inserting a deceiving class text onto the image in a "typographic" attack. These attacks succeed due to VLMs' text-heavy bias-a result of captions that echo visible words rather than describing content. However, this attack has focused solely on text that matches the target class exactly, overlooking a broader range of correlations, including non-matching text and graphical symbols, which arise from the abundance of branding content in web-scale data. To address this gap, we introduce "artifact-based" attacks: a novel class of manipulations that mislead models using both non-matching text and graphical elements. Unlike typographic attacks, these artifacts are not predefined, making them simultaneously harder to defend against and more challenging to find. We address this by framing artifact attacks as a search problem and demonstrate their effectiveness across five datasets, with some artifacts reinforcing each other to reach 100% attack success rates. These attacks transfer across models with up to 90% effectiveness, making it possible to attack unseen models. To defend against these attacks, we extend prior work's artifact aware prompting to the graphical setting. We see a moderate reduction of success rates of up to 15% relative to standard prompts, suggesting a promising direction for enhancing model robustness. Code: https://github.com/mqraitem/Web-Artifact-Attacks

Dante Francisco Wasmuht, Otto Brookes, Maximillian Schall et al.

Automated video analysis is critical for wildlife conservation. A foundational task in this domain is multi-animal tracking (MAT), which underpins applications such as individual re-identification and behavior recognition. However, existing datasets are limited in scale, constrained to a few species, or lack sufficient temporal and geographical diversity - leaving no suitable benchmark for training general-purpose MAT models applicable across wild animal populations. To address this, we introduce SA-FARI, the largest open-source MAT dataset for wild animals. It comprises 11,609 camera trap videos collected over approximately 10 years (2014-2024) from 741 locations across 4 continents, spanning 99 species categories. Each video is exhaustively annotated culminating in ~46 hours of densely annotated footage containing 16,224 masklet identities and 942,702 individual bounding boxes, segmentation masks, and species labels. Alongside the task-specific annotations, we publish anonymized camera trap locations for each video. Finally, we present comprehensive benchmarks on SA-FARI using state-of-the-art vision-language models for detection and tracking, including SAM 3, evaluated with both species-specific and generic animal prompts. We also compare against vision-only methods developed specifically for wildlife analysis. SA-FARI is the first large-scale dataset to combine high species diversity, multi-region coverage, and high-quality spatio-temporal annotations, offering a new foundation for advancing generalizable multianimal tracking in the wild. The dataset is available at $\href{https://www.conservationxlabs.com/sa-fari}{\text{conservationxlabs.com/SA-FARI}}$.

Samarth Mishra, Kate Saenko, Venkatesh Saligrama

Compositionality, or correctly recognizing scenes as compositions of atomic visual concepts, remains difficult for multimodal large language models (MLLMs). Even state of the art MLLMs such as GPT-4o can make mistakes in distinguishing compositions like "dog chasing cat" vs "cat chasing dog". While on Winoground, a benchmark for measuring such reasoning, MLLMs have made significant progress, they are still far from a human's performance. We show that compositional reasoning in these models can be improved by elucidating such concepts via data, where a model is trained to prefer the correct caption for an image over a close but incorrect one. We introduce SCRAMBLe: Synthetic Compositional Reasoning Augmentation of MLLMs with Binary preference Learning, an approach for preference tuning open-weight MLLMs on synthetic preference data generated in a fully automated manner from existing image-caption data. SCRAMBLe holistically improves these MLLMs' compositional reasoning capabilities which we can see through significant improvements across multiple vision language compositionality benchmarks, as well as smaller but significant improvements on general question answering tasks. As a sneak peek, SCRAMBLe tuned Molmo-7B model improves on Winoground from 49.5% to 54.8% (best reported to date), while improving by ~1% on more general visual question answering tasks. Code for SCRAMBLe along with tuned models and our synthetic training dataset is available at https://github.com/samarth4149/SCRAMBLe.

Shiori Sagawa, Pang Wei Koh, Tony Lee et al.

Machine learning systems deployed in the wild are often trained on a source distribution but deployed on a different target distribution. Unlabeled data can be a powerful point of leverage for mitigating these distribution shifts, as it is frequently much more available than labeled data and can often be obtained from distributions beyond the source distribution as well. However, existing distribution shift benchmarks with unlabeled data do not reflect the breadth of scenarios that arise in real-world applications. In this work, we present the WILDS 2.0 update, which extends 8 of the 10 datasets in the WILDS benchmark of distribution shifts to include curated unlabeled data that would be realistically obtainable in deployment. These datasets span a wide range of applications (from histology to wildlife conservation), tasks (classification, regression, and detection), and modalities (photos, satellite images, microscope slides, text, molecular graphs). The update maintains consistency with the original WILDS benchmark by using identical labeled training, validation, and test sets, as well as the evaluation metrics. On these datasets, we systematically benchmark state-of-the-art methods that leverage unlabeled data, including domain-invariant, self-training, and self-supervised methods, and show that their success on WILDS is limited. To facilitate method development and evaluation, we provide an open-source package that automates data loading and contains all of the model architectures and methods used in this paper. Code and leaderboards are available at https://wilds.stanford.edu.

Kuniaki Saito, Donghyun Kim, Piotr Teterwak et al.

Unsupervised domain adaptation (UDA) methods can dramatically improve generalization on unlabeled target domains. However, optimal hyper-parameter selection is critical to achieving high accuracy and avoiding negative transfer. Supervised hyper-parameter validation is not possible without labeled target data, which raises the question: How can we validate unsupervised adaptation techniques in a realistic way? We first empirically analyze existing criteria and demonstrate that they are not very effective for tuning hyper-parameters. Intuitively, a well-trained source classifier should embed target samples of the same class nearby, forming dense neighborhoods in feature space. Based on this assumption, we propose a novel unsupervised validation criterion that measures the density of soft neighborhoods by computing the entropy of the similarity distribution between points. Our criterion is simpler than competing validation methods, yet more effective; it can tune hyper-parameters and the number of training iterations in both image classification and semantic segmentation models. The code used for the paper will be available at \url{https://github.com/VisionLearningGroup/SND}.

Samarth Mishra, Kate Saenko, Venkatesh Saligrama

Most modern unsupervised domain adaptation (UDA) approaches are rooted in domain alignment, i.e., learning to align source and target features to learn a target domain classifier using source labels. In semi-supervised domain adaptation (SSDA), when the learner can access few target domain labels, prior approaches have followed UDA theory to use domain alignment for learning. We show that the case of SSDA is different and a good target classifier can be learned without needing alignment. We use self-supervised pretraining (via rotation prediction) and consistency regularization to achieve well separated target clusters, aiding in learning a low error target classifier. With our Pretraining and Consistency (PAC) approach, we achieve state of the art target accuracy on this semi-supervised domain adaptation task, surpassing multiple adversarial domain alignment methods, across multiple datasets. PAC, while using simple techniques, performs remarkably well on large and challenging SSDA benchmarks like DomainNet and Visda-17, often outperforming recent state of the art by sizeable margins. Code for our experiments can be found at https://github.com/venkatesh-saligrama/PAC

Kuniaki Saito, Donghyun Kim, Stan Sclaroff et al.

Unsupervised domain adaptation methods traditionally assume that all source categories are present in the target domain. In practice, little may be known about the category overlap between the two domains. While some methods address target settings with either partial or open-set categories, they assume that the particular setting is known a priori. We propose a more universally applicable domain adaptation framework that can handle arbitrary category shift, called Domain Adaptative Neighborhood Clustering via Entropy optimization (DANCE). DANCE combines two novel ideas: First, as we cannot fully rely on source categories to learn features discriminative for the target, we propose a novel neighborhood clustering technique to learn the structure of the target domain in a self-supervised way. Second, we use entropy-based feature alignment and rejection to align target features with the source, or reject them as unknown categories based on their entropy. We show through extensive experiments that DANCE outperforms baselines across open-set, open-partial and partial domain adaptation settings. Implementation is available at https://github.com/VisionLearningGroup/DANCE.

Bryan A. Plummer, Mariya I. Vasileva, Vitali Petsiuk et al.

Explaining a deep learning model can help users understand its behavior and allow researchers to discern its shortcomings. Recent work has primarily focused on explaining models for tasks like image classification or visual question answering. In this paper, we introduce Salient Attributes for Network Explanation (SANE) to explain image similarity models, where a model's output is a score measuring the similarity of two inputs rather than a classification score. In this task, an explanation depends on both of the input images, so standard methods do not apply. Our SANE explanations pairs a saliency map identifying important image regions with an attribute that best explains the match. We find that our explanations provide additional information not typically captured by saliency maps alone, and can also improve performance on the classic task of attribute recognition. Our approach's ability to generalize is demonstrated on two datasets from diverse domains, Polyvore Outfits and Animals with Attributes 2. Code available at: https://github.com/VisionLearningGroup/SANE

Ronghang Hu, Anna Rohrbach, Trevor Darrell et al.

Solving grounded language tasks often requires reasoning about relationships between objects in the context of a given task. For example, to answer the question "What color is the mug on the plate?" we must check the color of the specific mug that satisfies the "on" relationship with respect to the plate. Recent work has proposed various methods capable of complex relational reasoning. However, most of their power is in the inference structure, while the scene is represented with simple local appearance features. In this paper, we take an alternate approach and build contextualized representations for objects in a visual scene to support relational reasoning. We propose a general framework of Language-Conditioned Graph Networks (LCGN), where each node represents an object, and is described by a context-aware representation from related objects through iterative message passing conditioned on the textual input. E.g., conditioning on the "on" relationship to the plate, the object "mug" gathers messages from the object "plate" to update its representation to "mug on the plate", which can be easily consumed by a simple classifier for answer prediction. We experimentally show that our LCGN approach effectively supports relational reasoning and improves performance across several tasks and datasets. Our code is available at http://ronghanghu.com/lcgn.

Kuniaki Saito, Yoshitaka Ushiku, Tatsuya Harada et al.

We propose an approach for unsupervised adaptation of object detectors from label-rich to label-poor domains which can significantly reduce annotation costs associated with detection. Recently, approaches that align distributions of source and target images using an adversarial loss have been proven effective for adapting object classifiers. However, for object detection, fully matching the entire distributions of source and target images to each other at the global image level may fail, as domains could have distinct scene layouts and different combinations of objects. On the other hand, strong matching of local features such as texture and color makes sense, as it does not change category level semantics. This motivates us to propose a novel method for detector adaptation based on strong local alignment and weak global alignment. Our key contribution is the weak alignment model, which focuses the adversarial alignment loss on images that are globally similar and puts less emphasis on aligning images that are globally dissimilar. Additionally, we design the strong domain alignment model to only look at local receptive fields of the feature map. We empirically verify the effectiveness of our method on four datasets comprising both large and small domain shifts. Our code is available at \url{https://github.com/VisionLearningGroup/DA_Detection}

Ximeng Sun, Huijuan Xu, Kate Saenko

Video generation is an inherently challenging task, as it requires modeling realistic temporal dynamics as well as spatial content. Existing methods entangle the two intrinsically different tasks of motion and content creation in a single generator network, but this approach struggles to simultaneously generate plausible motion and content. To im-prove motion modeling in video generation tasks, we propose a two-stream model that disentangles motion generation from content generation, called a Two-Stream Variational Adversarial Network (TwoStreamVAN). Given an action label and a noise vector, our model is able to create clear and consistent motion, and thus yields photorealistic videos. The key idea is to progressively generate and fuse multi-scale motion with its corresponding spatial content. Our model significantly outperforms existing methods on the standard Weizmann Human Action, MUG Facial Expression, and VoxCeleb datasets, as well as our new dataset of diverse human actions with challenging and complex motion. Our code is available at https://github.com/sunxm2357/TwoStreamVAN/.

Huijuan Xu, Abir Das, Kate Saenko

We address the problem of activity detection in continuous, untrimmed video streams. This is a difficult task that requires extracting meaningful spatio-temporal features to capture activities, accurately localizing the start and end times of each activity. We introduce a new model, Region Convolutional 3D Network (R-C3D), which encodes the video streams using a three-dimensional fully convolutional network, then generates candidate temporal regions containing activities, and finally classifies selected regions into specific activities. Computation is saved due to the sharing of convolutional features between the proposal and the classification pipelines. The entire model is trained end-to-end with jointly optimized localization and classification losses. R-C3D is faster than existing methods (569 frames per second on a single Titan X Maxwell GPU) and achieves state-of-the-art results on THUMOS'14. We further demonstrate that our model is a general activity detection framework that does not rely on assumptions about particular dataset properties by evaluating our approach on ActivityNet and Charades. Our code is available at http://ai.bu.edu/r-c3d/.

Vasili Ramanishka, Abir Das, Jianming Zhang et al.

Neural image/video captioning models can generate accurate descriptions, but their internal process of mapping regions to words is a black box and therefore difficult to explain. Top-down neural saliency methods can find important regions given a high-level semantic task such as object classification, but cannot use a natural language sentence as the top-down input for the task. In this paper, we propose Caption-Guided Visual Saliency to expose the region-to-word mapping in modern encoder-decoder networks and demonstrate that it is learned implicitly from caption training data, without any pixel-level annotations. Our approach can produce spatial or spatiotemporal heatmaps for both predicted captions, and for arbitrary query sentences. It recovers saliency without the overhead of introducing explicit attention layers, and can be used to analyze a variety of existing model architectures and improve their design. Evaluation on large-scale video and image datasets demonstrates that our approach achieves comparable captioning performance with existing methods while providing more accurate saliency heatmaps. Our code is available at visionlearninggroup.github.io/caption-guided-saliency/.

Baochen Sun, Jiashi Feng, Kate Saenko

In this chapter, we present CORrelation ALignment (CORAL), a simple yet effective method for unsupervised domain adaptation. CORAL minimizes domain shift by aligning the second-order statistics of source and target distributions, without requiring any target labels. In contrast to subspace manifold methods, it aligns the original feature distributions of the source and target domains, rather than the bases of lower-dimensional subspaces. It is also much simpler than other distribution matching methods. CORAL performs remarkably well in extensive evaluations on standard benchmark datasets. We first describe a solution that applies a linear transformation to source features to align them with target features before classifier training. For linear classifiers, we propose to equivalently apply CORAL to the classifier weights, leading to added efficiency when the number of classifiers is small but the number and dimensionality of target examples are very high. The resulting CORAL Linear Discriminant Analysis (CORAL-LDA) outperforms LDA by a large margin on standard domain adaptation benchmarks. Finally, we extend CORAL to learn a nonlinear transformation that aligns correlations of layer activations in deep neural networks (DNNs). The resulting Deep CORAL approach works seamlessly with DNNs and achieves state-of-the-art performance on standard benchmark datasets. Our code is available at:~\url{https://github.com/VisionLearningGroup/CORAL}

Reuben Tan, Ximeng Sun, Ping Hu et al.

Long video question answering is a challenging task that involves recognizing short-term activities and reasoning about their fine-grained relationships. State-of-the-art video Large Language Models (vLLMs) hold promise as a viable solution due to their demonstrated emergent capabilities on new tasks. However, despite being trained on millions of short seconds-long videos, vLLMs are unable to understand minutes-long videos and accurately answer questions about them. To address this limitation, we propose a lightweight and self-supervised approach, Key frame-conditioned long video-LLM (Koala), that introduces learnable spatiotemporal queries to adapt pretrained vLLMs for generalizing to longer videos. Our approach introduces two new tokenizers that condition on visual tokens computed from sparse video key frames for understanding short and long video moments. We train our proposed approach on HowTo100M and demonstrate its effectiveness on zero-shot long video understanding benchmarks, where it outperforms state-of-the-art large models by 3 - 6% in absolute accuracy across all tasks. Surprisingly, we also empirically show that our approach not only helps a pretrained vLLM to understand long videos but also improves its accuracy on short-term action recognition.