Sujoy Paul

Mihir Prabhudesai, Anirudh Goyal, Sujoy Paul et al. · cmu

Current visual detectors, though impressive within their training distribution, often fail to parse out-of-distribution scenes into their constituent entities. Recent test-time adaptation methods use auxiliary self-supervised losses to adapt the network parameters to each test example independently and have shown promising results towards generalization outside the training distribution for the task of image classification. In our work, we find evidence that these losses are insufficient for the task of scene decomposition, without also considering architectural inductive biases. Recent slot-centric generative models attempt to decompose scenes into entities in a self-supervised manner by reconstructing pixels. Drawing upon these two lines of work, we propose Slot-TTA, a semi-supervised slot-centric scene decomposition model that at test time is adapted per scene through gradient descent on reconstruction or cross-view synthesis objectives. We evaluate Slot-TTA across multiple input modalities, images or 3D point clouds, and show substantial out-of-distribution performance improvements against state-of-the-art supervised feed-forward detectors, and alternative test-time adaptation methods.

Gagan Jain, Nidhi Hegde, Aditya Kusupati et al. · uw

The visual medium (images and videos) naturally contains a large amount of information redundancy, thereby providing a great opportunity for leveraging efficiency in processing. While Vision Transformer (ViT) based models scale effectively to large data regimes, they fail to capitalize on this inherent redundancy, leading to higher computational costs. Mixture of Experts (MoE) networks demonstrate scalability while maintaining same inference-time costs, but they come with a larger parameter footprint. We present Mixture of Nested Experts (MoNE), which utilizes a nested structure for experts, wherein individual experts fall on an increasing compute-accuracy curve. Given a compute budget, MoNE learns to dynamically choose tokens in a priority order, and thus redundant tokens are processed through cheaper nested experts. Using this framework, we achieve equivalent performance as the baseline models, while reducing inference time compute by over two-fold. We validate our approach on standard image and video datasets - ImageNet-21K, Kinetics400, and Something-Something-v2. We further highlight MoNE$'$s adaptability by showcasing its ability to maintain strong performance across different inference-time compute budgets on videos, using only a single trained model.

K J Joseph, Sujoy Paul, Gaurav Aggarwal et al.

Humans possess an innate ability to identify and differentiate instances that they are not familiar with, by leveraging and adapting the knowledge that they have acquired so far. Importantly, they achieve this without deteriorating the performance on their earlier learning. Inspired by this, we identify and formulate a new, pragmatic problem setting of NCDwF: Novel Class Discovery without Forgetting, which tasks a machine learning model to incrementally discover novel categories of instances from unlabeled data, while maintaining its performance on the previously seen categories. We propose 1) a method to generate pseudo-latent representations which act as a proxy for (no longer available) labeled data, thereby alleviating forgetting, 2) a mutual-information based regularizer which enhances unsupervised discovery of novel classes, and 3) a simple Known Class Identifier which aids generalized inference when the testing data contains instances form both seen and unseen categories. We introduce experimental protocols based on CIFAR-10, CIFAR-100 and ImageNet-1000 to measure the trade-off between knowledge retention and novel class discovery. Our extensive evaluations reveal that existing models catastrophically forget previously seen categories while identifying novel categories, while our method is able to effectively balance between the competing objectives. We hope our work will attract further research into this newly identified pragmatic problem setting.

K J Joseph, Sujoy Paul, Gaurav Aggarwal et al.

Novel Class Discovery (NCD) is a learning paradigm, where a machine learning model is tasked to semantically group instances from unlabeled data, by utilizing labeled instances from a disjoint set of classes. In this work, we first characterize existing NCD approaches into single-stage and two-stage methods based on whether they require access to labeled and unlabeled data together while discovering new classes. Next, we devise a simple yet powerful loss function that enforces separability in the latent space using cues from multi-dimensional scaling, which we refer to as Spacing Loss. Our proposed formulation can either operate as a standalone method or can be plugged into existing methods to enhance them. We validate the efficacy of Spacing Loss with thorough experimental evaluation across multiple settings on CIFAR-10 and CIFAR-100 datasets.

Nidhi Hegde, Sujoy Paul, Gagan Madan et al.

Recent document question answering models consist of two key components: the vision encoder, which captures layout and visual elements in images, and a Large Language Model (LLM) that helps contextualize questions to the image and supplements them with external world knowledge to generate accurate answers. However, the relative contributions of the vision encoder and the language model in these tasks remain unclear. This is especially interesting given the effectiveness of instruction-tuned LLMs, which exhibit remarkable adaptability to new tasks. To this end, we explore the following aspects in this work: (1) The efficacy of an LLM-only approach on document question answering tasks (2) strategies for serializing textual information within document images and feeding it directly to an instruction-tuned LLM, thus bypassing the need for an explicit vision encoder (3) thorough quantitative analysis on the feasibility of such an approach. Our comprehensive analysis encompasses six diverse benchmark datasets, utilizing LLMs of varying scales. Our findings reveal that a strategy exclusively reliant on the LLM yields results that are on par with or closely approach state-of-the-art performance across a range of datasets. We posit that this evaluation framework will serve as a guiding resource for selecting appropriate datasets for future research endeavors that emphasize the fundamental importance of layout and image content information.

Sujoy Paul, Gagan Madan, Akankshya Mishra et al.

State-of-the-art information extraction methods are limited by OCR errors. They work well for printed text in form-like documents, but unstructured, handwritten documents still remain a challenge. Adapting existing models to domain-specific training data is quite expensive, because of two factors, 1) limited availability of the domain-specific documents (such as handwritten prescriptions, lab notes, etc.), and 2) annotations become even more challenging as one needs domain-specific knowledge to decode inscrutable handwritten document images. In this work, we focus on the complex problem of extracting medicine names from handwritten prescriptions using only weakly labeled data. The data consists of images along with the list of medicine names in it, but not their location in the image. We solve the problem by first identifying the regions of interest, i.e., medicine lines from just weak labels and then injecting a domain-specific medicine language model learned using only synthetically generated data. Compared to off-the-shelf state-of-the-art methods, our approach performs >2.5x better in medicine names extraction from prescriptions.

Zahidul Islam, Sujoy Paul, Mrigank Rochan

With the exponential growth of video content, the need for automated video highlight detection to extract key moments or highlights from lengthy videos has become increasingly pressing. This technology has the potential to enhance user experiences by allowing quick access to relevant content across diverse domains. Existing methods typically rely either on expensive manually labeled frame-level annotations, or on a large external dataset of videos for weak supervision through category information. To overcome this, we focus on unsupervised video highlight detection, eliminating the need for manual annotations. We propose a novel unsupervised approach which capitalizes on the premise that significant moments tend to recur across multiple videos of the similar category in both audio and visual modalities. Surprisingly, audio remains under-explored, especially in unsupervised algorithms, despite its potential to detect key moments. Through a clustering technique, we identify pseudo-categories of videos and compute audio pseudo-highlight scores for each video by measuring the similarities of audio features among audio clips of all the videos within each pseudo-category. Similarly, we also compute visual pseudo-highlight scores for each video using visual features. Then, we combine audio and visual pseudo-highlights to create the audio-visual pseudo ground-truth highlight of each video for training an audio-visual highlight detection network. Extensive experiments and ablation studies on three benchmarks showcase the superior performance of our method over prior work.

Rajat Koner, Gagan Jain, Prateek Jain et al.

Vision Transformers (ViT) have emerged as the de-facto choice for numerous industry grade vision solutions. But their inference cost can be prohibitive for many settings, as they compute self-attention in each layer which suffers from quadratic computational complexity in the number of tokens. On the other hand, spatial information in images and spatio-temporal information in videos is usually sparse and redundant. In this work, we introduce LookupViT, that aims to exploit this information sparsity to reduce ViT inference cost. LookupViT provides a novel general purpose vision transformer block that operates by compressing information from higher resolution tokens to a fixed number of tokens. These few compressed tokens undergo meticulous processing, while the higher-resolution tokens are passed through computationally cheaper layers. Information sharing between these two token sets is enabled through a bidirectional cross-attention mechanism. The approach offers multiple advantages - (a) easy to implement on standard ML accelerators (GPUs/TPUs) via standard high-level operators, (b) applicable to standard ViT and its variants, thus generalizes to various tasks, (c) can handle different tokenization and attention approaches. LookupViT also offers flexibility for the compressed tokens, enabling performance-computation trade-offs in a single trained model. We show LookupViT's effectiveness on multiple domains - (a) for image-classification (ImageNet-1K and ImageNet-21K), (b) video classification (Kinetics400 and Something-Something V2), (c) image captioning (COCO-Captions) with a frozen encoder. LookupViT provides $2\times$ reduction in FLOPs while upholding or improving accuracy across these domains. In addition, LookupViT also demonstrates out-of-the-box robustness and generalization on image classification (ImageNet-C,R,A,O), improving by up to $4\%$ over ViT.

Debapriya Tula, Sujoy Paul, Gagan Madan et al.

Recognizing text lines from images is a challenging problem, especially for handwritten documents due to large variations in writing styles. While text line recognition models are generally trained on large corpora of real and synthetic data, such models can still make frequent mistakes if the handwriting is inscrutable or the image acquisition process adds corruptions, such as noise, blur, compression, etc. Writing style is generally quite consistent for an individual, which can be leveraged to correct mistakes made by such models. Motivated by this, we introduce the problem of adapting text line recognition models during test time. We focus on a challenging and realistic setting where, given only a single test image consisting of multiple text lines, the task is to adapt the model such that it performs better on the image, without any labels. We propose an iterative self-training approach that uses feedback from the language model to update the optical model, with confident self-labels in each iteration. The confidence measure is based on an augmentation mechanism that evaluates the divergence of the prediction of the model in a local region. We perform rigorous evaluation of our method on several benchmark datasets as well as their corrupted versions. Experimental results on multiple datasets spanning multiple scripts show that the proposed adaptation method offers an absolute improvement of up to 8% in character error rate with just a few iterations of self-training at test time.

Sahil Goyal, Swayam Agrawal, Gautham Govind Anil et al.

We introduce Elastic Looped Transformers (ELT), a highly parameter-efficient class of visual generative models based on a recurrent transformer architecture. While conventional generative models rely on deep stacks of unique transformer layers, our approach employs iterative, weight-shared transformer blocks to drastically reduce parameter counts while maintaining high synthesis quality. To effectively train these models for image and video generation, we propose the idea of Intra-Loop Self Distillation (ILSD), where student configurations (intermediate loops) are distilled from the teacher configuration (maximum training loops) to ensure consistency across the model's depth in a single training step. Our framework yields a family of elastic models from a single training run, enabling Any-Time inference capability with dynamic trade-offs between computational cost and generation quality, with the same parameter count. ELT significantly shifts the efficiency frontier for visual synthesis. With $4\times$ reduction in parameter count under iso-inference-compute settings, ELT achieves a competitive FID of $2.0$ on class-conditional ImageNet $256 \times 256$ and FVD of $72.8$ on class-conditional UCF-101.

Sahil Goyal, Debapriya Tula, Gagan Jain et al.

Recent advances in visual generation have made significant strides in producing content of exceptional quality. However, most methods suffer from a fundamental problem - a bottleneck of inference computational efficiency. Most of these algorithms involve multiple passes over a transformer model to generate tokens or denoise inputs. However, the model size is kept consistent throughout all iterations, which makes it computationally expensive. In this work, we aim to address this issue primarily through two key ideas - (a) not all parts of the generation process need equal compute, and we design a decode time model scaling schedule to utilize compute effectively, and (b) we can cache and reuse some of the computation. Combining these two ideas leads to using smaller models to process more tokens while large models process fewer tokens. These different-sized models do not increase the parameter size, as they share parameters. We rigorously experiment with ImageNet256$\times$256 , UCF101, and Kinetics600 to showcase the efficacy of the proposed method for image/video generation and frame prediction. Our experiments show that with almost $3\times$ less compute than baseline, our model obtains competitive performance.

Zahidul Islam, Sujoy Paul, Mrigank Rochan

Existing video highlight detection methods, although advanced, struggle to generalize well to all test videos. These methods typically employ a generic highlight detection model for each test video, which is suboptimal as it fails to account for the unique characteristics and variations of individual test videos. Such fixed models do not adapt to the diverse content, styles, or audio and visual qualities present in new, unseen test videos, leading to reduced highlight detection performance. In this paper, we propose Highlight-TTA, a test-time adaptation framework for video highlight detection that addresses this limitation by dynamically adapting the model during testing to better align with the specific characteristics of each test video, thereby improving generalization and highlight detection performance. Highlight-TTA is jointly optimized with an auxiliary task, cross-modality hallucinations, alongside the primary highlight detection task. We utilize a meta-auxiliary training scheme to enable effective adaptation through the auxiliary task while enhancing the primary task. During testing, we adapt the trained model using the auxiliary task on the test video to further enhance its highlight detection performance. Extensive experiments with three state-of-the-art highlight detection models and three benchmark datasets show that the introduction of Highlight-TTA to these models improves their performance, yielding superior results.

Sujoy Paul, Ansh Khurana, Gaurav Aggarwal

We consider the novel problem of unsupervised domain adaptation of source models, without access to the source data for semantic segmentation. Unsupervised domain adaptation aims to adapt a model learned on the labeled source data, to a new unlabeled target dataset. Existing methods assume that the source data is available along with the target data during adaptation. However, in practical scenarios, we may only have access to the source model and the unlabeled target data, but not the labeled source, due to reasons such as privacy, storage, etc. In this work, we propose a self-training approach to extract the knowledge from the source model. To compensate for the distribution shift from source to target, we first update only the normalization parameters of the network with the unlabeled target data. Then we employ confidence-filtered pseudo labeling and enforce consistencies against certain transformations. Despite being very simple and intuitive, our framework is able to achieve significant performance gains compared to directly applying the source model on the target data as reflected in our extensive experiments and ablation studies. In fact, the performance is just a few points away from the recent state-of-the-art methods which use source data for adaptation. We further demonstrate the generalisability of the proposed approach for fully test-time adaptation setting, where we do not need any target training data and adapt only during test-time.

Ansh Khurana, Sujoy Paul, Piyush Rai et al.

In Test-time Adaptation (TTA), given a source model, the goal is to adapt it to make better predictions for test instances from a different distribution than the source. Crucially, TTA assumes no access to the source data or even any additional labeled/unlabeled samples from the target distribution to finetune the source model. In this work, we consider TTA in a more pragmatic setting which we refer to as SITA (Single Image Test-time Adaptation). Here, when making a prediction, the model has access only to the given single test instance, rather than a batch of instances, as typically been considered in the literature. This is motivated by the realistic scenarios where inference is needed on-demand instead of delaying for an incoming batch or the inference is happening on an edge device (like mobile phone) where there is no scope for batching. The entire adaptation process in SITA should be extremely fast as it happens at inference time. To address this, we propose a novel approach AugBN that requires only a single forward pass. It can be used on any off-the-shelf trained model to test single instances for both classification and segmentation tasks. AugBN estimates normalization statistics of the unseen test distribution from the given test image using only one forward pass with label-preserving transformations. Since AugBN does not involve any back-propagation, it is significantly faster compared to recent test time adaptation methods. We further extend AugBN to make the algorithm hyperparameter-free. Rigorous experimentation show that our simple algorithm is able to achieve significant performance gains for a variety of datasets, tasks, and network architectures.

Sayak Nag, Dripta S. Raychaudhuri, Sujoy Paul et al.

In many applications, we are constrained to learn classifiers from very limited data (few-shot classification). The task becomes even more challenging if it is also required to identify samples from unknown categories (open-set classification). Learning a good abstraction for a class with very few samples is extremely difficult, especially under open-set settings. As a result, open-set recognition has received minimal attention in the few-shot setting. However, it is a critical task in many applications like environmental monitoring, where the number of labeled examples for each class is limited. Existing few-shot open-set recognition (FSOSR) methods rely on thresholding schemes, with some considering uniform probability for open-class samples. However, this approach is often inaccurate, especially for fine-grained categorization, and makes them highly sensitive to the choice of a threshold. To address these concerns, we propose Reconstructing Exemplar-based Few-shot Open-set ClaSsifier (ReFOCS). By using a novel exemplar reconstruction-based meta-learning strategy ReFOCS streamlines FSOSR eliminating the need for a carefully tuned threshold by learning to be self-aware of the openness of a sample. The exemplars, act as class representatives and can be either provided in the training dataset or estimated in the feature domain. By testing on a wide variety of datasets, we show ReFOCS to outperform multiple state-of-the-art methods.

Dripta S. Raychaudhuri, Sujoy Paul, Jeroen van Baar et al.

Imitation learning seeks to circumvent the difficulty in designing proper reward functions for training agents by utilizing expert behavior. With environments modeled as Markov Decision Processes (MDP), most of the existing imitation algorithms are contingent on the availability of expert demonstrations in the same MDP as the one in which a new imitation policy is to be learned. In this paper, we study the problem of how to imitate tasks when there exist discrepancies between the expert and agent MDP. These discrepancies across domains could include differing dynamics, viewpoint, or morphology; we present a novel framework to learn correspondences across such domains. Importantly, in contrast to prior works, we use unpaired and unaligned trajectories containing only states in the expert domain, to learn this correspondence. We utilize a cycle-consistency constraint on both the state space and a domain agnostic latent space to do this. In addition, we enforce consistency on the temporal position of states via a normalized position estimator function, to align the trajectories across the two domains. Once this correspondence is found, we can directly transfer the demonstrations on one domain to the other and use it for imitation. Experiments across a wide variety of challenging domains demonstrate the efficacy of our approach.

Sk Miraj Ahmed, Dripta S. Raychaudhuri, Sujoy Paul et al.

Unsupervised Domain Adaptation (UDA) aims to learn a predictor model for an unlabeled domain by transferring knowledge from a separate labeled source domain. However, most of these conventional UDA approaches make the strong assumption of having access to the source data during training, which may not be very practical due to privacy, security and storage concerns. A recent line of work addressed this problem and proposed an algorithm that transfers knowledge to the unlabeled target domain from a single source model without requiring access to the source data. However, for adaptation purposes, if there are multiple trained source models available to choose from, this method has to go through adapting each and every model individually, to check for the best source. Thus, we ask the question: can we find the optimal combination of source models, with no source data and without target labels, whose performance is no worse than the single best source? To answer this, we propose a novel and efficient algorithm which automatically combines the source models with suitable weights in such a way that it performs at least as good as the best source model. We provide intuitive theoretical insights to justify our claim. Furthermore, extensive experiments are conducted on several benchmark datasets to show the effectiveness of our algorithm, where in most cases, our method not only reaches best source accuracy but also outperforms it.

Akash Gupta, Rameswar Panda, Sujoy Paul et al.

While machine learning approaches to visual recognition offer great promise, most of the existing methods rely heavily on the availability of large quantities of labeled training data. However, in the vast majority of real-world settings, manually collecting such large labeled datasets is infeasible due to the cost of labeling data or the paucity of data in a given domain. In this paper, we present a novel Adversarial Knowledge Transfer (AKT) framework for transferring knowledge from internet-scale unlabeled data to improve the performance of a classifier on a given visual recognition task. The proposed adversarial learning framework aligns the feature space of the unlabeled source data with the labeled target data such that the target classifier can be used to predict pseudo labels on the source data. An important novel aspect of our method is that the unlabeled source data can be of different classes from those of the labeled target data, and there is no need to define a separate pretext task, unlike some existing approaches. Extensive experiments well demonstrate that models learned using our approach hold a lot of promise across a variety of visual recognition tasks on multiple standard datasets.

Sujoy Paul, Yi-Hsuan Tsai, Samuel Schulter et al.

Learning semantic segmentation models requires a huge amount of pixel-wise labeling. However, labeled data may only be available abundantly in a domain different from the desired target domain, which only has minimal or no annotations. In this work, we propose a novel framework for domain adaptation in semantic segmentation with image-level weak labels in the target domain. The weak labels may be obtained based on a model prediction for unsupervised domain adaptation (UDA), or from a human annotator in a new weakly-supervised domain adaptation (WDA) paradigm for semantic segmentation. Using weak labels is both practical and useful, since (i) collecting image-level target annotations is comparably cheap in WDA and incurs no cost in UDA, and (ii) it opens the opportunity for category-wise domain alignment. Our framework uses weak labels to enable the interplay between feature alignment and pseudo-labeling, improving both in the process of domain adaptation. Specifically, we develop a weak-label classification module to enforce the network to attend to certain categories, and then use such training signals to guide the proposed category-wise alignment method. In experiments, we show considerable improvements with respect to the existing state-of-the-arts in UDA and present a new benchmark in the WDA setting. Project page is at http://www.nec-labs.com/~mas/WeakSegDA.

Xueping Wang, Sujoy Paul, Dripta S. Raychaudhuri et al.

Most person re-identification methods, being supervised techniques, suffer from the burden of massive annotation requirement. Unsupervised methods overcome this need for labeled data, but perform poorly compared to the supervised alternatives. In order to cope with this issue, we introduce the problem of learning person re-identification models from videos with weak supervision. The weak nature of the supervision arises from the requirement of video-level labels, i.e. person identities who appear in the video, in contrast to the more precise framelevel annotations. Towards this goal, we propose a multiple instance attention learning framework for person re-identification using such video-level labels. Specifically, we first cast the video person re-identification task into a multiple instance learning setting, in which person images in a video are collected into a bag. The relations between videos with similar labels can be utilized to identify persons, on top of that, we introduce a co-person attention mechanism which mines the similarity correlations between videos with person identities in common. The attention weights are obtained based on all person images instead of person tracklets in a video, making our learned model less affected by noisy annotations. Extensive experiments demonstrate the superiority of the proposed method over the related methods on two weakly labeled person re-identification datasets.

Sujoy Paul, Jeroen van Baar, Amit K. Roy-Chowdhury

Learning to solve complex goal-oriented tasks with sparse terminal-only rewards often requires an enormous number of samples. In such cases, using a set of expert trajectories could help to learn faster. However, Imitation Learning (IL) via supervised pre-training with these trajectories may not perform as well and generally requires additional finetuning with expert-in-the-loop. In this paper, we propose an approach which uses the expert trajectories and learns to decompose the complex main task into smaller sub-goals. We learn a function which partitions the state-space into sub-goals, which can then be used to design an extrinsic reward function. We follow a strategy where the agent first learns from the trajectories using IL and then switches to Reinforcement Learning (RL) using the identified sub-goals, to alleviate the errors in the IL step. To deal with states which are under-represented by the trajectory set, we also learn a function to modulate the sub-goal predictions. We show that our method is able to solve complex goal-oriented tasks, which other RL, IL or their combinations in literature are not able to solve.

Mahmudul Hasan, Sujoy Paul, Anastasios I. Mourikis et al.

Activity recognition is a challenging problem with many practical applications. In addition to the visual features, recent approaches have benefited from the use of context, e.g., inter-relationships among the activities and objects. However, these approaches require data to be labeled, entirely available beforehand, and not designed to be updated continuously, which make them unsuitable for surveillance applications. In contrast, we propose a continuous-learning framework for context-aware activity recognition from unlabeled video, which has two distinct advantages over existing methods. First, it employs a novel active-learning technique that not only exploits the informativeness of the individual activities but also utilizes their contextual information during query selection; this leads to significant reduction in expensive manual annotation effort. Second, the learned models can be adapted online as more data is available. We formulate a conditional random field model that encodes the context and devise an information-theoretic approach that utilizes entropy and mutual information of the nodes to compute the set of most informative queries, which are labeled by a human. These labels are combined with graphical inference techniques for incremental updates. We provide a theoretical formulation of the active learning framework with an analytic solution. Experiments on six challenging datasets demonstrate that our framework achieves superior performance with significantly less manual labeling.

Niluthpol Chowdhury Mithun, Sujoy Paul, Amit K. Roy-Chowdhury

There have been a few recent methods proposed in text to video moment retrieval using natural language queries, but requiring full supervision during training. However, acquiring a large number of training videos with temporal boundary annotations for each text description is extremely time-consuming and often not scalable. In order to cope with this issue, in this work, we introduce the problem of learning from weak labels for the task of text to video moment retrieval. The weak nature of the supervision is because, during training, we only have access to the video-text pairs rather than the temporal extent of the video to which different text descriptions relate. We propose a joint visual-semantic embedding based framework that learns the notion of relevant segments from video using only video-level sentence descriptions. Specifically, our main idea is to utilize latent alignment between video frames and sentence descriptions using Text-Guided Attention (TGA). TGA is then used during the test phase to retrieve relevant moments. Experiments on two benchmark datasets demonstrate that our method achieves comparable performance to state-of-the-art fully supervised approaches.

Sujoy Paul, Jeroen van Baar

Deep Reinforcement Learning has shown tremendous success in solving several games and tasks in robotics. However, unlike humans, it generally requires a lot of training instances. Trajectories imitating to solve the task at hand can help to increase sample-efficiency of deep RL methods. In this paper, we present a simple approach to use such trajectories, applied to the challenging Ball-in-Maze Games, recently introduced in the literature. We show that in spite of not using human-generated trajectories and just using the simulator as a model to generate a limited number of trajectories, we can get a speed-up of about 2-3x in the learning process. We also discuss some challenges we observed while using trajectory-based learning for very sparse reward functions.

Sujoy Paul, Sourya Roy, Amit K. Roy-Chowdhury

Deep neural networks are efficient learning machines which leverage upon a large amount of manually labeled data for learning discriminative features. However, acquiring substantial amount of supervised data, especially for videos can be a tedious job across various computer vision tasks. This necessitates learning of visual features from videos in an unsupervised setting. In this paper, we propose a computationally simple, yet effective, framework to learn spatio-temporal feature embedding from unlabeled videos. We train a Convolutional 3D Siamese network using positive and negative pairs mined from videos under certain probabilistic assumptions. Experimental results on three datasets demonstrate that our proposed framework is able to learn weights which can be used for same as well as cross dataset and tasks.

Sujoy Paul, Sourya Roy, Amit K Roy-Chowdhury

Most activity localization methods in the literature suffer from the burden of frame-wise annotation requirement. Learning from weak labels may be a potential solution towards reducing such manual labeling effort. Recent years have witnessed a substantial influx of tagged videos on the Internet, which can serve as a rich source of weakly-supervised training data. Specifically, the correlations between videos with similar tags can be utilized to temporally localize the activities. Towards this goal, we present W-TALC, a Weakly-supervised Temporal Activity Localization and Classification framework using only video-level labels. The proposed network can be divided into two sub-networks, namely the Two-Stream based feature extractor network and a weakly-supervised module, which we learn by optimizing two complimentary loss functions. Qualitative and quantitative results on two challenging datasets - Thumos14 and ActivityNet1.2, demonstrate that the proposed method is able to detect activities at a fine granularity and achieve better performance than current state-of-the-art methods.

Shasha Li, Ajaya Neupane, Sujoy Paul et al.

Recent research has demonstrated the brittleness of machine learning systems to adversarial perturbations. However, the studies have been mostly limited to perturbations on images and more generally, classification that does not deal with temporally varying inputs. In this paper we ask "Are adversarial perturbations possible in real-time video classification systems and if so, what properties must they satisfy?" Such systems find application in surveillance applications, smart vehicles, and smart elderly care and thus, misclassification could be particularly harmful (e.g., a mishap at an elderly care facility may be missed). We show that accounting for temporal structure is key to generating adversarial examples in such systems. We exploit recent advances in generative adversarial network (GAN) architectures to account for temporal correlations and generate adversarial samples that can cause misclassification rates of over 80% for targeted activities. More importantly, the samples also leave other activities largely unaffected making them extremely stealthy. Finally, we also surprisingly find that in many scenarios, the same perturbation can be applied to every frame in a video clip that makes the adversary's ability to achieve misclassification relatively easy.