Laura Sevilla-Lara

Shreyank N Gowda, Marcus Rohrbach, Frank Keller et al.

We address the problem of data augmentation for video action recognition. Standard augmentation strategies in video are hand-designed and sample the space of possible augmented data points either at random, without knowing which augmented points will be better, or through heuristics. We propose to learn what makes a good video for action recognition and select only high-quality samples for augmentation. In particular, we choose video compositing of a foreground and a background video as the data augmentation process, which results in diverse and realistic new samples. We learn which pairs of videos to augment without having to actually composite them. This reduces the space of possible augmentations, which has two advantages: it saves computational cost and increases the accuracy of the final trained classifier, as the augmented pairs are of higher quality than average. We present experimental results on the entire spectrum of training settings: few-shot, semi-supervised and fully supervised. We observe consistent improvements across all of them over prior work and baselines on Kinetics, UCF101, HMDB51, and achieve a new state-of-the-art on settings with limited data. We see improvements of up to 8.6% in the semi-supervised setting.

Gen Li, Varun Jampani, Deqing Sun et al.

Humans excel at acquiring knowledge through observation. For example, we can learn to use new tools by watching demonstrations. This skill is fundamental for intelligent systems to interact with the world. A key step to acquire this skill is to identify what part of the object affords each action, which is called affordance grounding. In this paper, we address this problem and propose a framework called LOCATE that can identify matching object parts across images, to transfer knowledge from images where an object is being used (exocentric images used for learning), to images where the object is inactive (egocentric ones used to test). To this end, we first find interaction areas and extract their feature embeddings. Then we learn to aggregate the embeddings into compact prototypes (human, object part, and background), and select the one representing the object part. Finally, we use the selected prototype to guide affordance grounding. We do this in a weakly supervised manner, learning only from image-level affordance and object labels. Extensive experiments demonstrate that our approach outperforms state-of-the-art methods by a large margin on both seen and unseen objects.

Kiyoon Kim, Davide Moltisanti, Oisin Mac Aodha et al.

Precisely naming the action depicted in a video can be a challenging and oftentimes ambiguous task. In contrast to object instances represented as nouns (e.g. dog, cat, chair, etc.), in the case of actions, human annotators typically lack a consensus as to what constitutes a specific action (e.g. jogging versus running). In practice, a given video can contain multiple valid positive annotations for the same action. As a result, video datasets often contain significant levels of label noise and overlap between the atomic action classes. In this work, we address the challenge of training multi-label action recognition models from only single positive training labels. We propose two approaches that are based on generating pseudo training examples sampled from similar instances within the train set. Unlike other approaches that use model-derived pseudo-labels, our pseudo-labels come from human annotations and are selected based on feature similarity. To validate our approaches, we create a new evaluation benchmark by manually annotating a subset of EPIC-Kitchens-100's validation set with multiple verb labels. We present results on this new test set along with additional results on a new version of HMDB-51, called Confusing-HMDB-102, where we outperform existing methods in both cases. Data and code are available at https://github.com/kiyoon/verb_ambiguity

Shreyank N Gowda, Laura Sevilla-Lara

Video understanding has long suffered from reliance on large labeled datasets, motivating research into zero-shot learning. Recent progress in language modeling presents opportunities to advance zero-shot video analysis, but constructing an effective semantic space relating action classes remains challenging. We address this by introducing a novel dataset, Stories, which contains rich textual descriptions for diverse action classes extracted from WikiHow articles. For each class, we extract multi-sentence narratives detailing the necessary steps, scenes, objects, and verbs that characterize the action. This contextual data enables modeling of nuanced relationships between actions, paving the way for zero-shot transfer. We also propose an approach that harnesses Stories to improve feature generation for training zero-shot classification. Without any target dataset fine-tuning, our method achieves new state-of-the-art on multiple benchmarks, improving top-1 accuracy by up to 6.1%. We believe Stories provides a valuable resource that can catalyze progress in zero-shot action recognition. The textual narratives forge connections between seen and unseen classes, overcoming the bottleneck of labeled data that has long impeded advancements in this exciting domain. The data can be found here: https://github.com/kini5gowda/Stories .

Gen Li, Nikolaos Tsagkas, Jifei Song et al.

Affordance, defined as the potential actions that an object offers, is crucial for embodied AI agents. For example, such knowledge directs an agent to grasp a knife by the handle for cutting or by the blade for safe handover. While existing approaches have made notable progress, affordance research still faces three key challenges: data scarcity, poor generalization, and real-world deployment. Specifically, there is a lack of large-scale affordance datasets with precise segmentation maps, existing models struggle to generalize across different domains or novel object and affordance classes, and little work demonstrates deployability in real-world scenarios. In this work, we address these issues by proposing a complete affordance learning system that (1) takes in egocentric videos and outputs precise affordance annotations without human labeling, (2) leverages geometric information and vision foundation models to improve generalization, and (3) introduces a framework that facilitates affordance-oriented robotic manipulation such as tool grasping and robot-to-human tool handover. Experimental results show that our model surpasses the state-of-the-art by 13.8% in mIoU, and the framework achieves 77.1% successful grasping among 179 trials, including evaluations on seen, unseen classes, and cluttered scenes. Project page: https://reagan1311.github.io/affgrasp.

Gen Li, Deqing Sun, Laura Sevilla-Lara et al.

We introduce One-shot Open Affordance Learning (OOAL), where a model is trained with just one example per base object category, but is expected to identify novel objects and affordances. While vision-language models excel at recognizing novel objects and scenes, they often struggle to understand finer levels of granularity such as affordances. To handle this issue, we conduct a comprehensive analysis of existing foundation models, to explore their inherent understanding of affordances and assess the potential for data-limited affordance learning. We then propose a vision-language framework with simple and effective designs that boost the alignment between visual features and affordance text embeddings. Experiments on two affordance segmentation benchmarks show that the proposed method outperforms state-of-the-art models with less than 1% of the full training data, and exhibits reasonable generalization capability on unseen objects and affordances.

Madhav Agarwal, Sotirios A. Tsaftaris, Laura Sevilla-Lara et al.

Understanding emotions is a fundamental ability for intelligent systems to be able to interact with humans. Vision-language models (VLMs) have made tremendous progress in the last few years for many visual tasks, potentially offering a promising solution for understanding emotions. However, it is surprising that even the most sophisticated contemporary VLMs struggle to recognize human emotions or to outperform even specialized vision-only classifiers. In this paper we ask the question "Why do VLMs struggle to recognize human emotions?", and observe that the inherently continuous and dynamic task of facial expression recognition (DFER) exposes two critical VLM vulnerabilities. First, emotion datasets are naturally long-tailed, and the web-scale data used to pre-train VLMs exacerbates this head-class bias, causing them to systematically collapse rare, under-represented emotions into common categories. We propose alternative sampling strategies that prevent favoring common concepts. Second, temporal information is critical for understanding emotions. However, VLMs are unable to represent temporal information over dense frame sequences, as they are limited by context size and the number of tokens that can fit in memory, which poses a clear challenge for emotion recognition. We demonstrate that the sparse temporal sampling strategy used in VLMs is inherently misaligned with the fleeting nature of micro-expressions (0.25-0.5 seconds), which are often the most critical affective signal. As a diagnostic probe, we propose a multi-stage context enrichment strategy that utilizes the information from "in-between" frames by first converting them into natural language summaries. This enriched textual context is provided as input to the VLM alongside sparse keyframes, preventing attentional dilution from excessive visual data while preserving the emotional trajectory.

Davide Moltisanti, Frank Keller, Hakan Bilen et al.

The goal of this work is to understand the way actions are performed in videos. That is, given a video, we aim to predict an adverb indicating a modification applied to the action (e.g. cut "finely"). We cast this problem as a regression task. We measure textual relationships between verbs and adverbs to generate a regression target representing the action change we aim to learn. We test our approach on a range of datasets and achieve state-of-the-art results on both adverb prediction and antonym classification. Furthermore, we outperform previous work when we lift two commonly assumed conditions: the availability of action labels during testing and the pairing of adverbs as antonyms. Existing datasets for adverb recognition are either noisy, which makes learning difficult, or contain actions whose appearance is not influenced by adverbs, which makes evaluation less reliable. To address this, we collect a new high quality dataset: Adverbs in Recipes (AIR). We focus on instructional recipes videos, curating a set of actions that exhibit meaningful visual changes when performed differently. Videos in AIR are more tightly trimmed and were manually reviewed by multiple annotators to ensure high labelling quality. Results show that models learn better from AIR given its cleaner videos. At the same time, adverb prediction on AIR is challenging, demonstrating that there is considerable room for improvement.

Shreyank N Gowda, Xinyue Hao, Gen Li et al.

Deep learning models have revolutionized various fields, from image recognition to natural language processing, by achieving unprecedented levels of accuracy. However, their increasing energy consumption has raised concerns about their environmental impact, disadvantaging smaller entities in research and exacerbating global energy consumption. In this paper, we explore the trade-off between model accuracy and electricity consumption, proposing a metric that penalizes large consumption of electricity. We conduct a comprehensive study on the electricity consumption of various deep learning models across different GPUs, presenting a detailed analysis of their accuracy-efficiency trade-offs. By evaluating accuracy per unit of electricity consumed, we demonstrate how smaller, more energy-efficient models can significantly expedite research while mitigating environmental concerns. Our results highlight the potential for a more sustainable approach to deep learning, emphasizing the importance of optimizing models for efficiency. This research also contributes to a more equitable research landscape, where smaller entities can compete effectively with larger counterparts. This advocates for the adoption of efficient deep learning practices to reduce electricity consumption, safeguarding the environment for future generations whilst also helping ensure a fairer competitive landscape.

Anil Batra, Shreyank N Gowda, Frank Keller et al.

Understanding the steps required to perform a task is an important skill for AI systems. Learning these steps from instructional videos involves two subproblems: (i) identifying the temporal boundary of sequentially occurring segments and (ii) summarizing these steps in natural language. We refer to this task as Procedure Segmentation and Summarization (PSS). In this paper, we take a closer look at PSS and propose three fundamental improvements over current methods. The segmentation task is critical, as generating a correct summary requires each step of the procedure to be correctly identified. However, current segmentation metrics often overestimate the segmentation quality because they do not consider the temporal order of segments. In our first contribution, we propose a new segmentation metric that takes into account the order of segments, giving a more reliable measure of the accuracy of a given predicted segmentation. Current PSS methods are typically trained by proposing segments, matching them with the ground truth and computing a loss. However, much like segmentation metrics, existing matching algorithms do not consider the temporal order of the mapping between candidate segments and the ground truth. In our second contribution, we propose a matching algorithm that constrains the temporal order of segment mapping, and is also differentiable. Lastly, we introduce multi-modal feature training for PSS, which further improves segmentation. We evaluate our approach on two instructional video datasets (YouCook2 and Tasty) and observe an improvement over the state-of-the-art of $\sim7\%$ and $\sim2.5\%$ for procedure segmentation and summarization, respectively.

Mantas Skackauskas, Xinyue Hao, Laura Sevilla-Lara

Video representation learning has seen tremendous progress in recent years. This has been driven by many factors, including the scale of training and the success of visual models trained contrastively with language. While these factors have pushed the boundaries of what video models can do, they also introduce their own set of limitations: first, scaling video models can reach prohibitive costs and second, learning from language restricts the range of concepts that can be learned to those in captions. As a result, video models still struggle with temporal understanding. In this paper we propose a novel approach that uses motion as the central modality for video representation. In particular, given the motion in a video in the form of point-tracks, we use a masked-autoencoder to mask some of the tracks and train the autoencoder to reconstruct the missing tracks. This allows us to learn a representation in a self-supervised manner. We show that using motion to represent videos actually addresses both of the core limitations of video technology. First, it allows us to massively reduce the scale of training data, as motion is inherently appearance-independent and hence needs fewer examples to generalize well. Second, motion allows us to bypass the language-dependent training paradigm, learning better fine-grained concepts. The result is an embedding that we call TIME (Temporally Informed Motion Embedding), a representation trained exclusively on synthetic motion data. We test this embedding on a wide set of tasks in a zero-shot manner. We observe that without bells and whistles, performance is on par with state-of-the-art models using up to 4 orders of magnitude less training data. This is a stepping stone towards a new paradigm of video models that are both more temporally aware as well as more scalable.

Madhav Agarwal, Mingtian Zhang, Laura Sevilla-Lara et al.

Speech-driven talking heads have recently emerged and enable interactive avatars. However, real-world applications are limited, as current methods achieve high visual fidelity but slow or fast yet temporally unstable. Diffusion methods provide realistic image generation, yet struggle with oneshot settings. Gaussian Splatting approaches are real-time, yet inaccuracies in facial tracking, or inconsistent Gaussian mappings, lead to unstable outputs and video artifacts that are detrimental to realistic use cases. We address this problem by mapping Gaussian Splatting using 3D Morphable Models to generate person-specific avatars. We introduce transformer-based prediction of model parameters, directly from audio, to drive temporal consistency. From monocular video and independent audio speech inputs, our method enables generation of real-time talking head videos where we report competitive quantitative and qualitative performance.

Anil Batra, Davide Moltisanti, Laura Sevilla-Lara et al.

Procedural videos, exemplified by recipe demonstrations, are instrumental in conveying step-by-step instructions. However, understanding such videos is challenging as it involves the precise localization of steps and the generation of textual instructions. Manually annotating steps and writing instructions is costly, which limits the size of current datasets and hinders effective learning. Leveraging large but noisy video-transcript datasets for pre-training can boost performance but demands significant computational resources. Furthermore, transcripts contain irrelevant content and differ in style from human-written instructions. To mitigate these issues, we propose a novel technique, Sieve-&-Swap, to automatically generate high-quality training data for the recipe domain: (i) Sieve: filters irrelevant transcripts and (ii) Swap: acquires high-quality text by replacing transcripts with human-written instruction from a text-only recipe dataset. The resulting dataset is three orders of magnitude smaller than current web-scale datasets but enables efficient training of large-scale models. Alongside Sieve-&-Swap, we propose Procedure Transformer (ProcX), a model for end-to-end step localization and instruction generation for procedural videos. When pre-trained on our curated dataset, this model achieves state-of-the-art performance on YouCook2 and Tasty while using a fraction of the training data. We have released code and dataset.

Shriram M Sathiyanarayanan, Xinyue Hao, Shihao Hou et al.

The success of the machine learning field has reliably depended on training on large datasets. While effective, this trend comes at an extraordinary cost. This is due to two deeply intertwined factors: the size of models and the size of datasets. While promising research efforts focus on reducing the size of models, the other half of the equation remains fairly mysterious. Indeed, it is surprising that the standard approach to training remains to iterate over and over, uniformly sampling the training dataset. In this paper we explore a series of alternative training paradigms that leverage insights from hard-data-mining and dropout, simple enough to implement and use that can become the new training standard. The proposed Progressive Data Dropout reduces the number of effective epochs to as little as 12.4% of the baseline. This savings actually do not come at any cost for accuracy. Surprisingly, the proposed method improves accuracy by up to 4.82%. Our approach requires no changes to model architecture or optimizer, and can be applied across standard training pipelines, thus posing an excellent opportunity for wide adoption. Code can be found here: https://github.com/bazyagami/LearningWithRevision

Kiyoon Kim, Shreyank N Gowda, Oisin Mac Aodha et al.

We address the problem of capturing temporal information for video classification in 2D networks, without increasing their computational cost. Existing approaches focus on modifying the architecture of 2D networks (e.g. by including filters in the temporal dimension to turn them into 3D networks, or using optical flow, etc.), which increases computation cost. Instead, we propose a novel sampling strategy, where we re-order the channels of the input video, to capture short-term frame-to-frame changes. We observe that without bells and whistles, the proposed sampling strategy improves performance on multiple architectures (e.g. TSN, TRN, TSM, and MVFNet) and datasets (CATER, Something-Something-V1 and V2), up to 24% over the baseline of using the standard video input. In addition, our sampling strategies do not require training from scratch and do not increase the computational cost of training and testing. Given the generality of the results and the flexibility of the approach, we hope this can be widely useful to the video understanding community. Code is available on our website: https://github.com/kiyoon/channel_sampling.

Shreyank N Gowda, Davide Moltisanti, Laura Sevilla-Lara

Zero-shot action recognition requires a strong ability to generalize from pre-training and seen classes to novel unseen classes. Similarly, continual learning aims to develop models that can generalize effectively and learn new tasks without forgetting the ones previously learned. The generalization goals of zero-shot and continual learning are closely aligned, however techniques from continual learning have not been applied to zero-shot action recognition. In this paper, we propose a novel method based on continual learning to address zero-shot action recognition. This model, which we call {\em Generative Iterative Learning} (GIL) uses a memory of synthesized features of past classes, and combines these synthetic features with real ones from novel classes. The memory is used to train a classification model, ensuring a balanced exposure to both old and new classes. Experiments demonstrate that {\em GIL} improves generalization in unseen classes, achieving a new state-of-the-art in zero-shot recognition across multiple benchmarks. Importantly, {\em GIL} also boosts performance in the more challenging generalized zero-shot setting, where models need to retain knowledge about classes seen before fine-tuning.

Willem Davison, Xinyue Hao, Laura Sevilla-Lara

Temporal information has long been considered to be essential for perception. While there is extensive research on the role of image information for perceptual tasks, the role of the temporal dimension remains less well understood: What can we learn about the world from long-term motion information? What properties does long-term motion information have for visual learning? We leverage recent success in point-track estimation, which offers an excellent opportunity to learn temporal representations and experiment on a variety of perceptual tasks. We draw 3 clear lessons: 1) Long-term motion representations contain information to understand actions, but also objects, materials, and spatial information, often even better than images. 2) Long-term motion representations generalize far better than image representations in low-data settings and in zero-shot tasks. 3) The very low dimensionality of motion information makes motion representations a better trade-off between GFLOPs and accuracy than standard video representations, and used together they achieve higher performance than video representations alone. We hope these insights will pave the way for the design of future models that leverage the power of long-term motion information for perception.

Gen Li, Bo Zhao, Jianfei Yang et al.

Generating interaction-centric videos, such as those depicting humans or robots interacting with objects, is crucial for embodied intelligence, as they provide rich and diverse visual priors for robot learning, manipulation policy training, and affordance reasoning. However, existing methods often struggle to model such complex and dynamic interactions. While recent studies show that masks can serve as effective control signals and enhance generation quality, obtaining dense and precise mask annotations remains a major challenge for real-world use. To overcome this limitation, we introduce Mask2IV, a novel framework specifically designed for interaction-centric video generation. It adopts a decoupled two-stage pipeline that first predicts plausible motion trajectories for both actor and object, then generates a video conditioned on these trajectories. This design eliminates the need for dense mask inputs from users while preserving the flexibility to manipulate the interaction process. Furthermore, Mask2IV supports versatile and intuitive control, allowing users to specify the target object of interaction and guide the motion trajectory through action descriptions or spatial position cues. To support systematic training and evaluation, we curate two benchmarks covering diverse action and object categories across both human-object interaction and robotic manipulation scenarios. Extensive experiments demonstrate that our method achieves superior visual realism and controllability compared to existing baselines.

Anil Batra, Laura Sevilla-Lara, Marcus Rohrbach et al.

Procedural texts help AI enhance reasoning about context and action sequences. Transforming these into Semantic Role Labeling (SRL) improves understanding of individual steps by identifying predicate-argument structure like {verb,what,where/with}. Procedural instructions are highly elliptic, for instance, (i) add cucumber to the bowl and (ii) add sliced tomatoes, the second step's where argument is inferred from the context, referring to where the cucumber was placed. Prior SRL benchmarks often miss implicit arguments, leading to incomplete understanding. To address this, we introduce Implicit-VidSRL, a dataset that necessitates inferring implicit and explicit arguments from contextual information in multimodal cooking procedures. Our proposed dataset benchmarks multimodal models' contextual reasoning, requiring entity tracking through visual changes in recipes. We study recent multimodal LLMs and reveal that they struggle to predict implicit arguments of what and where/with from multi-modal procedural data given the verb. Lastly, we propose iSRL-Qwen2-VL, which achieves a 17% relative improvement in F1-score for what-implicit and a 14.7% for where/with-implicit semantic roles over GPT-4o.

Xinyue Hao, Gen Li, Shreyank N Gowda et al.

Video understanding has made huge strides in recent years, relying largely on the power of transformers. As this architecture is notoriously expensive and video data is highly redundant, research into improving efficiency has become particularly relevant. Some creative solutions include token selection and merging. While most methods succeed in reducing the cost of the model and maintaining accuracy, an interesting pattern arises: most methods do not outperform the baseline of randomly discarding tokens. In this paper we take a closer look at this phenomenon and observe 5 principles of the nature of visual tokens. For example, we observe that the value of tokens follows a clear Pareto-distribution where most tokens have remarkably low value, and just a few carry most of the perceptual information. We build on these and further insights to propose a lightweight video model, LITE, that can select a small number of tokens effectively, outperforming state-of-the-art and existing baselines across datasets (Kinetics-400 and Something-Something-V2) in the challenging trade-off of computation (GFLOPs) vs accuracy. Experiments also show that LITE generalizes across datasets and even other tasks without the need for retraining.

Davide Moltisanti, Hakan Bilen, Laura Sevilla-Lara et al.

We focus on the problem of recognising the end state of an action in an image, which is critical for understanding what action is performed and in which manner. We study this focusing on the task of predicting the coarseness of a cut, i.e., deciding whether an object was cut "coarsely" or "finely". No dataset with these annotated end states is available, so we propose an augmentation method to synthesise training data. We apply this method to cutting actions extracted from an existing action recognition dataset. Our method is object agnostic, i.e., it presupposes the location of the object but not its identity. Starting from less than a hundred images of a whole object, we can generate several thousands images simulating visually diverse cuts of different coarseness. We use our synthetic data to train a model based on UNet and test it on real images showing coarsely/finely cut objects. Results demonstrate that the model successfully recognises the end state of the cutting action despite the domain gap between training and testing, and that the model generalises well to unseen objects.

Shreyank N Gowda, Laura Sevilla-Lara, Kiyoon Kim et al.

Zero-shot action recognition is the task of classifying action categories that are not available in the training set. In this setting, the standard evaluation protocol is to use existing action recognition datasets(e.g. UCF101) and randomly split the classes into seen and unseen. However, most recent work builds on representations pre-trained on the Kinetics dataset, where classes largely overlap with classes in the zero-shot evaluation datasets. As a result, classes which are supposed to be unseen, are present during supervised pre-training, invalidating the condition of the zero-shot setting. A similar concern was previously noted several years ago for image based zero-shot recognition but has not been considered by the zero-shot action recognition community. In this paper, we propose a new split for true zero-shot action recognition with no overlap between unseen test classes and training or pre-training classes. We benchmark several recent approaches on the proposed True Zero-Shot(TruZe) Split for UCF101 and HMDB51, with zero-shot and generalized zero-shot evaluation. In our extensive analysis, we find that our TruZesplits are significantly harder than comparable random splits as nothing is leaking from pre-training, i.e. unseen performance is consistently lower,up to 8.9% for zero-shot action recognition. In an additional evaluation we also find that similar issues exist in the splits used in few-shot action recognition, here we see differences of up to 17.1%. We publish oursplits1and hope that our benchmark analysis will change how the field is evaluating zero- and few-shot action recognition moving forward.

Gen Li, Varun Jampani, Laura Sevilla-Lara et al.

Prototype learning is extensively used for few-shot segmentation. Typically, a single prototype is obtained from the support feature by averaging the global object information. However, using one prototype to represent all the information may lead to ambiguities. In this paper, we propose two novel modules, named superpixel-guided clustering (SGC) and guided prototype allocation (GPA), for multiple prototype extraction and allocation. Specifically, SGC is a parameter-free and training-free approach, which extracts more representative prototypes by aggregating similar feature vectors, while GPA is able to select matched prototypes to provide more accurate guidance. By integrating the SGC and GPA together, we propose the Adaptive Superpixel-guided Network (ASGNet), which is a lightweight model and adapts to object scale and shape variation. In addition, our network can easily generalize to k-shot segmentation with substantial improvement and no additional computational cost. In particular, our evaluations on COCO demonstrate that ASGNet surpasses the state-of-the-art method by 5% in 5-shot segmentation.

Shreyank N Gowda, Laura Sevilla-Lara, Frank Keller et al.

Zero-shot action recognition is the task of recognizingaction classes without visual examples, only with a seman-tic embedding which relates unseen to seen classes. Theproblem can be seen as learning a function which general-izes well to instances of unseen classes without losing dis-crimination between classes. Neural networks can modelthe complex boundaries between visual classes, which ex-plains their success as supervised models. However, inzero-shot learning, these highly specialized class bound-aries may not transfer well from seen to unseen classes.In this paper we propose a centroid-based representation,which clusters visual and semantic representation, consid-ers all training samples at once, and in this way generaliz-ing well to instances from unseen classes. We optimize theclustering using Reinforcement Learning which we show iscritical for our approach to work. We call the proposedmethod CLASTER and observe that it consistently outper-forms the state-of-the-art in all standard datasets, includ-ing UCF101, HMDB51 and Olympic Sports; both in thestandard zero-shot evaluation and the generalized zero-shotlearning. Further, we show that our model performs com-petitively in the image domain as well, outperforming thestate-of-the-art in many settings.

Shreyank N Gowda, Marcus Rohrbach, Laura Sevilla-Lara

Action recognition is computationally expensive. In this paper, we address the problem of frame selection to improve the accuracy of action recognition. In particular, we show that selecting good frames helps in action recognition performance even in the trimmed videos domain. Recent work has successfully leveraged frame selection for long, untrimmed videos, where much of the content is not relevant, and easy to discard. In this work, however, we focus on the more standard short, trimmed action recognition problem. We argue that good frame selection can not only reduce the computational cost of action recognition but also increase the accuracy by getting rid of frames that are hard to classify. In contrast to previous work, we propose a method that instead of selecting frames by considering one at a time, considers them jointly. This results in a more efficient selection, where good frames are more effectively distributed over the video, like snapshots that tell a story. We call the proposed frame selection SMART and we test it in combination with different backbone architectures and on multiple benchmarks (Kinetics, Something-something, UCF101). We show that the SMART frame selection consistently improves the accuracy compared to other frame selection strategies while reducing the computational cost by a factor of 4 to 10 times. Additionally, we show that when the primary goal is recognition performance, our selection strategy can improve over recent state-of-the-art models and frame selection strategies on various benchmarks (UCF101, HMDB51, FCVID, and ActivityNet).

Shreyank N Gowda, Panagiotis Eustratiadis, Timothy Hospedales et al.

We consider the challenging problem of zero-shot video object segmentation (VOS). That is, segmenting and tracking multiple moving objects within a video fully automatically, without any manual initialization. We treat this as a grouping problem by exploiting object proposals and making a joint inference about grouping over both space and time. We propose a network architecture for tractably performing proposal selection and joint grouping. Crucially, we then show how to train this network with reinforcement learning so that it learns to perform the optimal non-myopic sequence of grouping decisions to segment the whole video. Unlike standard supervised techniques, this also enables us to directly optimize for the non-differentiable overlap-based metrics used to evaluate VOS. We show that the proposed method, which we call ALBA outperforms the previous stateof-the-art on three benchmarks: DAVIS 2017 [2], FBMS [20] and Youtube-VOS [27].

Yannis Kalantidis, Laura Sevilla-Lara, Ernest Mwebaze et al.

This is the proceedings of the Computer Vision for Agriculture (CV4A) Workshop that was held in conjunction with the International Conference on Learning Representations (ICLR) 2020. The Computer Vision for Agriculture (CV4A) 2020 workshop was scheduled to be held in Addis Ababa, Ethiopia, on April 26th, 2020. It was held virtually that same day due to the COVID-19 pandemic. The workshop was held in conjunction with the International Conference on Learning Representations (ICLR) 2020.

Laura Sevilla-Lara, Shengxin Zha, Zhicheng Yan et al.

Understanding temporal information and how the visual world changes over time is a fundamental ability of intelligent systems. In video understanding, temporal information is at the core of many current challenges, including compression, efficient inference, motion estimation or summarization. However, in current video datasets it has been observed that action classes can often be recognized without any temporal information from a single frame of video. As a result, both benchmarking and training in these datasets may give an unintentional advantage to models with strong image understanding capabilities, as opposed to those with strong temporal understanding. In this paper we address this problem head on by identifying action classes where temporal information is actually necessary to recognize them and call these "temporal classes". Selecting temporal classes using a computational method would bias the process. Instead, we propose a methodology based on a simple and effective human annotation experiment. We remove just the temporal information by shuffling frames in time and measure if the action can still be recognized. Classes that cannot be recognized when frames are not in order are included in the temporal Dataset. We observe that this set is statistically different from other static classes, and that performance in it correlates with a network's ability to capture temporal information. Thus we use it as a benchmark on current popular networks, which reveals a series of interesting facts. We also explore the effect of training on the temporal dataset, and observe that this leads to better generalization in unseen classes, demonstrating the need for more temporal data. We hope that the proposed dataset of temporal categories will help guide future research in temporal modeling for better video understanding.

Linchao Zhu, Laura Sevilla-Lara, Du Tran et al.

Typical video classification methods often divide a video into short clips, do inference on each clip independently, then aggregate the clip-level predictions to generate the video-level results. However, processing visually similar clips independently ignores the temporal structure of the video sequence, and increases the computational cost at inference time. In this paper, we propose a novel framework named FASTER, i.e., Feature Aggregation for Spatio-TEmporal Redundancy. FASTER aims to leverage the redundancy between neighboring clips and reduce the computational cost by learning to aggregate the predictions from models of different complexities. The FASTER framework can integrate high quality representations from expensive models to capture subtle motion information and lightweight representations from cheap models to cover scene changes in the video. A new recurrent network (i.e., FAST-GRU) is designed to aggregate the mixture of different representations. Compared with existing approaches, FASTER can reduce the FLOPs by over 10x? while maintaining the state-of-the-art accuracy across popular datasets, such as Kinetics, UCF-101 and HMDB-51.

Zheng Shou, Xudong Lin, Yannis Kalantidis et al.

Motion has shown to be useful for video understanding, where motion is typically represented by optical flow. However, computing flow from video frames is very time-consuming. Recent works directly leverage the motion vectors and residuals readily available in the compressed video to represent motion at no cost. While this avoids flow computation, it also hurts accuracy since the motion vector is noisy and has substantially reduced resolution, which makes it a less discriminative motion representation. To remedy these issues, we propose a lightweight generator network, which reduces noises in motion vectors and captures fine motion details, achieving a more Discriminative Motion Cue (DMC) representation. Since optical flow is a more accurate motion representation, we train the DMC generator to approximate flow using a reconstruction loss and a generative adversarial loss, jointly with the downstream action classification task. Extensive evaluations on three action recognition benchmarks (HMDB-51, UCF-101, and a subset of Kinetics) confirm the effectiveness of our method. Our full system, consisting of the generator and the classifier, is coined as DMC-Net which obtains high accuracy close to that of using flow and runs two orders of magnitude faster than using optical flow at inference time.

Laura Sevilla-Lara, Yiyi Liao, Fatma Guney et al.

Most of the top performing action recognition methods use optical flow as a "black box" input. Here we take a deeper look at the combination of flow and action recognition, and investigate why optical flow is helpful, what makes a flow method good for action recognition, and how we can make it better. In particular, we investigate the impact of different flow algorithms and input transformations to better understand how these affect a state-of-the-art action recognition method. Furthermore, we fine tune two neural-network flow methods end-to-end on the most widely used action recognition dataset (UCF101). Based on these experiments, we make the following five observations: 1) optical flow is useful for action recognition because it is invariant to appearance, 2) optical flow methods are optimized to minimize end-point-error (EPE), but the EPE of current methods is not well correlated with action recognition performance, 3) for the flow methods tested, accuracy at boundaries and at small displacements is most correlated with action recognition performance, 4) training optical flow to minimize classification error instead of minimizing EPE improves recognition performance, and 5) optical flow learned for the task of action recognition differs from traditional optical flow especially inside the human body and at the boundary of the body. These observations may encourage optical flow researchers to look beyond EPE as a goal and guide action recognition researchers to seek better motion cues, leading to a tighter integration of the optical flow and action recognition communities.

Jonas Wulff, Laura Sevilla-Lara, Michael J. Black

The optical flow of natural scenes is a combination of the motion of the observer and the independent motion of objects. Existing algorithms typically focus on either recovering motion and structure under the assumption of a purely static world or optical flow for general unconstrained scenes. We combine these approaches in an optical flow algorithm that estimates an explicit segmentation of moving objects from appearance and physical constraints. In static regions we take advantage of strong constraints to jointly estimate the camera motion and the 3D structure of the scene over multiple frames. This allows us to also regularize the structure instead of the motion. Our formulation uses a Plane+Parallax framework, which works even under small baselines, and reduces the motion estimation to a one-dimensional search problem, resulting in more accurate estimation. In moving regions the flow is treated as unconstrained, and computed with an existing optical flow method. The resulting Mostly-Rigid Flow (MR-Flow) method achieves state-of-the-art results on both the MPI-Sintel and KITTI-2015 benchmarks.

Laura Sevilla-Lara, Deqing Sun, Varun Jampani et al.

Existing optical flow methods make generic, spatially homogeneous, assumptions about the spatial structure of the flow. In reality, optical flow varies across an image depending on object class. Simply put, different objects move differently. Here we exploit recent advances in static semantic scene segmentation to segment the image into objects of different types. We define different models of image motion in these regions depending on the type of object. For example, we model the motion on roads with homographies, vegetation with spatially smooth flow, and independently moving objects like cars and planes with affine motion plus deviations. We then pose the flow estimation problem using a novel formulation of localized layers, which addresses limitations of traditional layered models for dealing with complex scene motion. Our semantic flow method achieves the lowest error of any published monocular method in the KITTI-2015 flow benchmark and produces qualitatively better flow and segmentation than recent top methods on a wide range of natural videos.