Theory-of-mind (ToM) enables humans to infer mental states—such as beliefs, desires, and intentions—forming the foundation of social cognition. Existing computational ToM methods rely on structured workflows with ToM-specific priors or deep model fine-tuning but struggle with scalability in multimodal environments. They remain trapped within the gravitational pull of multi-step planning complexity, failing to generalize as task demands increase. To overcome these limitations, we propose a scalable Bayesian ToM planner. It breaks down ToM complexity into stepwise Bayesian updates. Meanwhile, weak-to-strong control specializes smaller LMs to refine ToM-specific likelihood estimation, transferring their ToM reasoning behavior to larger LMs (7B to 405B) for social and world knowledge integration. This synergistic approach enables scalability, aligning large-model inference with human mental states with Bayesian principles. Extensive experiments demonstrate a 4.6% improvement in accuracy over state-of-the-art methods on multimodal ToM benchmarks, including unseen scenarios, establishing a new standard for modeling human mental states in complex environments.

Vision-language models (VLMs) are capable of recognizing unseen actions. However, existing VLMs lack intrinsic understanding of procedural action concepts. Hence, they overfit to fixed labels and are not invariant to unseen action synonyms. To address this, we propose a simple fine-tuning technique, Action Concept Enhancement (ACE), to improve the robustness and concept understanding of VLMs in procedural action classification. ACE continually incorporates augmented action synonyms and negatives in an auxiliary classification loss by stochastically replacing fixed labels during training. This creates new combinations of action labels over the course of fine-tuning and prevents overfitting to fixed action representations. We show the enhanced concept understanding of our VLM, by visualizing the alignment of encoded embeddings of unseen action synonyms in the embedding space. Our experiments on the ATA, IKEA and GTEA datasets demonstrate the efficacy of ACE in domains of cooking and assembly leading to significant improvements in zero-shot action classification while maintaining competitive performance on seen actions.

Human action anticipation aims at predicting what people will do in the future based on past observations. In this paper, we introduce Uncertainty-aware Action Decoupling Transformer (UADT) for action anticipation. Unlike existing methods that directly predict action in a verb-noun pair format, we decouple the action anticipation task into verb and noun anticipations separately. The objective is to make the two decoupled tasks assist each other and eventually im-prove the action anticipation task. Specifically, we propose a two-stream Transformer-based architecture which is composed of a verb-to-noun model and a noun-to-verb model. The verb-to-noun model leverages the verb information to improve the noun prediction and the other way around. We extend the model in a probabilistic manner and quantify the predictive uncertainty of each decoupled task to select features. In this way, the noun prediction leverages the most in-formative and redundancy-free verb features and verb pre-diction works similarly. Finally, the two streams are combined dynamically based on their uncertainties to make the joint action anticipation. We demonstrate the efficacy of our method by achieving state-of-the-art performance on action anticipation benchmarks including EPIC-KITCHENS, EGTEA Gaze+, and 50-Salads.

We introduce PlausiVL, a large video-language model for anticipating action sequences that are plausible in the real-world. While significant efforts have been made towards anticipating future actions, prior approaches do not take into account the aspect of plausibility in an action sequence. To address this limitation, we explore the generative capability of a large video-language model in our work and further, develop the understanding of plausibility in an action sequence by introducing two objective functions, a counterfactual-based plausible action sequence learning loss and a long-horizon action repetition loss. We utilize temporal logical constraints as well as verb-noun action pair logical constraints to create implausible/counterfactual action sequences and use them to train the model with plausible action sequence learning loss. This loss helps the model to differentiate between plausible and not plausible action sequences and also helps the model to learn implicit temporal cues crucial for the task of action anticipation. The long-horizon action repetition loss puts a higher penalty on the actions that are more prone to repetition over a longer temporal window. With this penalization, the model is able to generate diverse, plausible action sequences. We evaluate our approach on two large-scale datasets, Ego4D and EPIC-Kitchens-100, and show improvements on the task of action anticipation.

Can we better anticipate an actor’s future actions (e.g. mix eggs) by knowing what commonly happens after the current action (e.g. crack eggs)? What if the actor also shares the goal (e.g. make fried rice) with us? The long-term action anticipation (LTA) task aims to predict an actor’s future behavior from video observations in the form of verb and noun sequences, and it is crucial for human-machine interaction. We propose to formulate the LTA task from two perspectives: a bottom-up approach that predicts the next actions autoregressively by modeling temporal dynamics; and a top-down approach that infers the goal of the actor and plans the needed procedure to accomplish the goal. We hypothesize that large language models (LLMs), which have been pretrained on procedure text data (e.g. recipes, how-tos), have the potential to help LTA from both perspectives. It can help provide the prior knowledge on the possible next actions, and infer the goal given the observed part of a procedure, respectively. We propose AntGPT, which represents video observations as sequences of human actions, and uses the action representation for an LLM to infer the goals and model temporal dynamics. AntGPT achieves state- of-the-art performance on Ego4D LTA v1 and v2, EPIC-Kitchens-55, as well as EGTEA GAZE+, thanks to LLMs’ goal inference and temporal dynamics modeling capabilities. We further demonstrate that these capabilities can be effectively distilled into a compact neural network 1.3% of the original LLM model size. Code and model will be released upon acceptance.

Real-world recommender systems often need to balance multiple objectives when deciding which recommendations to present to users. These include behavioural signals (e.g. clicks, shares, dwell time), as well as broader objectives (e.g. diversity, fairness). Scalarisation methods are commonly used to handle this balancing task, where a weighted average of per-objective reward signals determines the final score used for ranking. Naturally, how these weights are computed exactly, is key to success for any online platform.

This paper focuses on building object-centric representations for long-term action anticipation in videos. Our key motivation is that objects provide important cues to recognize and predict human-object interactions, especially when the predictions are longer term, as an observed "background" object could be used by the human actor in the future. We observe that existing object-based video recognition frameworks either assume the existence of in-domain supervised object detectors or follow a fully weakly-supervised pipeline to infer object locations from action labels. We propose to build object-centric video representations by leveraging visual-language pretrained models. This is achieved by "object prompts", an approach to extract task-specific object-centric representations from general-purpose pretrained models without finetuning. To recognize and predict human-object interactions, we use a Transformer-based neural architecture which allows the "retrieval" of relevant objects for action anticipation at various time scales. We conduct extensive evaluations on the Ego4D, 50Salads, and EGTEA Gaze+ benchmarks. Both quantitative and qualitative results confirm the effectiveness of our proposed method. Our code is available at github.com/brown-palm/ObjectPrompt.

The widespread adoption of commercial autonomous vehicles (AVs) and advanced driver assistance systems (ADAS) may largely depend on their acceptance by society, for which their perceived trustworthiness and interpretability to riders are crucial. In general, this task is challenging because modern autonomous systems software relies heavily on black-box artificial intelligence models. Towards this goal, this paper introduces a novel dataset, Rank2Tell1, a multi-modal ego-centric dataset for Ranking the importance level and Telling the reason for the importance. Using various close and open-ended visual question answering, the dataset provides dense annotations of various semantic, spatial, temporal, and relational attributes of various important objects in complex traffic scenarios. The dense annotations and unique attributes of the dataset make it a valuable resource for researchers working on visual scene understanding and related fields. Furthermore, we introduce a joint model for joint importance level ranking and natural language captions generation to benchmark our dataset and demonstrate performance with quantitative evaluations.

Predicting future action locations is vital for applications like human-robot collaboration. While some computer vision tasks have made progress in predicting human actions, accurately localizing these actions in future frames remains an area with room for improvement. We introduce a new task called spatial action localization in the future (SALF), which aims to predict action locations in both observed and future frames. SALF is challenging because it requires understanding the underlying physics of video observations to predict future action locations accurately. To address SALF, we use the concept of NeuralODE, which models the latent dynamics of sequential data by solving ordinary differential equations (ODE) with neural networks. We propose a novel architecture, AdamsFormer, which extends observed frame features to future time horizons by modeling continuous temporal dynamics through ODE solving. Specifically, we employ the Adams method, a multi-step approach that efficiently uses information from previous steps without discarding it. Our extensive experiments on UCF101-24 and JHMDB-21 datasets demonstrate that our proposed model outperforms existing long-range temporal modeling methods by a significant margin in terms of frame-mAP.

We present RAFTformer, a realtime action forecasting transformer for latency-aware realworld action forecasting. RAFTformer is a two-stage fully transformer based architecture comprising of a video transformer backbone that operates on high resolution, short-range clips, and a head transformer encoder that temporally aggregates information from multiple short-range clips to span a long-term horizon. Additionally, we propose a novel self-supervised shuffled causal masking scheme as a model level augmentation to improve forecasting fidelity. Finally, we also propose a novel realtime evaluation setting for action fore-casting that directly couples model inference latency to overall forecasting performance and brings forth a hith-erto overlooked trade-off between latency and action fore-casting performance. Our parsimonious network design fa-cilitates RAFTformer inference latency to be 9x smaller than prior works at the same forecasting accuracy. Owing to its two-staged design, RAFTformer uses 94% less training compute and 90% lesser training parameters to outperform prior state-of-the-art baselines by 4.9 points on EGTEA Gaze+ and by 1.4 points on EPIC-Kitchens-100 validation set, as measured by Top-5 recall (T5R) in the offline setting. In the realtime setting, RAFTformer outperforms prior works by an even greater margin of upto 4.4 T5R points on the EPIC-Kitchens-100 dataset. Project Webpage: https://karttikeya.github.io/publication/RAFTformer/.

We introduce a novel representation called Ordered Atomic Activity for interactive scenario understanding. The representation decomposes each scenario into a set of ordered atomic activities, where each activity consists of an action and the corresponding actors involved and the order denotes the temporal development of the scenario. This design also helps in identifying important interactive relationships, such as yielding. The action is a high-level semantic motion pattern that is grounded in the surrounding road topology, which we decompose into zones and corners with unique IDs. For example, a group of pedestrians crossing in front is denoted as C1 → C4: P+, as depicted in Figure 1. We collect a new large-scale dataset called OATS1(Ordered Atomic Activities in interactive Traffic Scenarios), comprising 1026 video clips (~ 20s) captured at intersections in San Francisco Bay Area. Each clip is labeled with the proposed language, resulting in 59 activity categories and 6512 annotated activity instances. We propose three fine-grained scenario understanding tasks, i.e., multilabel atomic activity recognition, activity order prediction, and interactive scenario retrieval. We also propose a Graph Convolutional Network based framework that models both appearance and motion of traffic participants to tackle the above tasks, that performs favorably against state-of-the-art methods. However, we find that the methods cannot achieve satisfactory performance, indicating rising opportunities for the community to develop new algorithms for these tasks towards better interactive scenario understanding.

We present a novel method for weakly-supervised action segmentation and unseen error detection in anomalous instructional videos. In the absence of an appropriate dataset for this task, we introduce the Anomalous Toy Assembly (ATA) dataset1, which comprises 1152 untrimmed videos of 32 participants assembling three different toys, recorded from four different viewpoints. The training set comprises 27 participants who assemble toys in an expected and consistent manner, while the test and validation sets comprise 5 participants who display sequential anomalies in their task. We introduce a weakly labeled segmentation algorithm that is a generalization of the constrained Viterbi algorithm and identifies potential anomalous moments based on the difference between future anticipation and current recognition results. The proposed method is not restricted by the training transcripts during testing, allowing for the inference of anomalous action sequences while maintaining real-time performance. Based on these segmentation results, we also introduce a baseline to detect pre-defined human errors, and benchmark results on the ATA dataset. Experiments were conducted on the ATA and CSV datasets, outperforming the state-of-the-art in segmenting anomalous videos under both online and offline conditions.

This paper addresses a new problem of weakly-supervised online action segmentation in instructional videos. We present a framework to segment streaming videos online at test time using Dynamic Programming and show its advantages over greedy sliding window approach. We improve our framework by introducing the Online-Offline Discrepancy Loss (OODL) to encourage the segmentation results to have a higher temporal consistency. Furthermore, only during training, we exploit framewise correspondence between multiple views as supervision for training weakly-labeled instructional videos. In particular, we investigate three different multi-view inference techniques to generate more accurate frame-wise pseudo ground-truth with no additional annotation cost. We present results and ablation studies on two benchmark multi-view datasets, Breakfast and IKEA ASM. Experimental results show efficacy of the proposed methods both qualitatively and quantitatively in two domains of cooking and assembly.