Toward Cognitive Supersensing in Multimodal Large Language Model

Multimodal Large Language Models (MLLMs) have achieved remarkable success in open-vocabulary perceptual tasks, yet their ability to solve complex cognitive problems remains limited, especially when visual details are abstract and require visual memory. Current approaches primarily scale Chain-of-Thought (CoT) reasoning in the text space, even when language alone is insufficient for clear and structured reasoning, and largely neglect visual reasoning mechanisms analogous to the human visuospatial sketchpad and visual imagery. To mitigate this deficiency, we introduce Cognitive Supersensing, a novel training paradigm that endows MLLMs with human-like visual imagery capabilities by integrating a Latent Visual Imagery Prediction (LVIP) head that jointly learns sequences of visual cognitive latent embeddings and aligns them with the answer, thereby forming vision-based internal reasoning chains. We further introduce a reinforcement learning stage that optimizes text reasoning paths based on this grounded visual latent. To evaluate the cognitive capabilities of MLLMs, we present CogSense-Bench, a comprehensive visual question answering (VQA) benchmark assessing five cognitive dimensions. Extensive experiments demonstrate that MLLMs trained with Cognitive Supersensing significantly outperform state-of-the-art baselines on CogSense-Bench and exhibit superior generalization on out-of-domain mathematics and science VQA benchmarks, suggesting that internal visual imagery is potentially key to bridging the gap between perceptual recognition and cognitive understanding. We will open-source the CogSense-Bench and our model weights.

How Much 3D Do Video Foundation Models Encode?

Videos are continuous 2D projections of 3D worlds. After training on large video data, will global 3D understanding naturally emerge? We study this by quantifying the 3D understanding of existing Video Foundation Models (VidFMs) pretrained on vast video data. We propose the first model-agnostic framework that measures the 3D awareness of various VidFMs by estimating multiple 3D properties from their features via shallow read-outs. Our study presents meaningful findings regarding the 3D awareness of VidFMs on multiple axes. In particular, we show that state-of-the-art video generation models exhibit a strong understanding of 3D objects and scenes, despite not being trained on any 3D data. Such understanding can even surpass that of large expert models specifically trained for 3D tasks. Our findings, together with the 3D benchmarking of major VidFMs, provide valuable observations for building scalable 3D models.

Improving Personalized Search with Regularized Low-Rank Parameter Updates

Personalized vision-language retrieval seeks to recognize new concepts (e.g. "my dog Fido") from only a few examples. This task is challenging because it requires not only learning a new concept from a few images, but also integrating the personal and general knowledge together to recognize the concept in different contexts. In this paper, we show how to effectively adapt the internal representation of a vision-language dual encoder model for personalized vision-language retrieval. We find that regularized low-rank adaption of a small set of parameters in the language encoder's final layer serves as a highly effective alternative to textual inversion for recognizing the personal concept while preserving general knowledge. Additionally, we explore strategies for combining parameters of multiple learned personal concepts, finding that parameter addition is effective. To evaluate how well general knowledge is preserved in a finetuned representation, we introduce a metric that measures image retrieval accuracy based on captions generated by a vision language model (VLM). Our approach achieves state-of-the-art accuracy on two benchmarks for personalized image retrieval with natural language queries - DeepFashion2 and ConCon-Chi - outperforming the prior art by 4%-22% on personal retrievals.

LSM-2: Learning from Incomplete Wearable Sensor Data

Foundation models, a cornerstone of recent advancements in machine learning, have predominantly thrived on complete and well-structured data. Wearable sensor data frequently suffers from significant missingness, posing a substantial challenge for self-supervised learning (SSL) models that typically assume complete data inputs. This paper introduces the second generation of Large Sensor Model (LSM-2) with Adaptive and Inherited Masking (AIM), a novel SSL approach that learns robust representations directly from incomplete data without requiring explicit imputation. AIM's core novelty lies in its use of learnable mask tokens to model both existing ("inherited") and artificially introduced missingness, enabling it to robustly handle fragmented real-world data during inference. Pre-trained on an extensive dataset of 40M hours of day-long multimodal sensor data, our LSM-2 with AIM achieves the best performance across a diverse range of tasks, including classification, regression and generative modeling. Furthermore, LSM-2 with AIM exhibits superior scaling performance, and critically, maintains high performance even under targeted missingness scenarios, reflecting clinically coherent patterns, such as the diagnostic value of nighttime biosignals for hypertension prediction. This makes AIM a more reliable choice for real-world wearable data applications.

Incorporating Flexible Image Conditioning into Text-to-Video Diffusion Models without Training

Text-image-to-video (TI2V) generation is a critical problem for controllable video generation using both semantic and visual conditions. Most existing methods typically add visual conditions to text-to-video (T2V) foundation models by finetuning, which is costly in resources and only limited to a few predefined conditioning settings. To tackle this issue, we introduce a unified formulation for TI2V generation with flexible visual conditioning. Furthermore, we propose an innovative training-free approach, dubbed FlexTI2V, that can condition T2V foundation models on an arbitrary amount of images at arbitrary positions. Specifically, we firstly invert the condition images to noisy representation in a latent space. Then, in the denoising process of T2V models, our method uses a novel random patch swapping strategy to incorporate visual features into video representations through local image patches. To balance creativity and fidelity, we use a dynamic control mechanism to adjust the strength of visual conditioning to each video frame. Extensive experiments validate that our method surpasses previous training-free image conditioning methods by a notable margin. We also show more insights of our method by detailed ablation study and analysis.

MEBench: A Novel Benchmark for Understanding Mutual Exclusivity Bias in Vision-Language Models

This paper introduces MEBench, a novel benchmark for evaluating mutual exclusivity (ME) bias, a cognitive phenomenon observed in children during word learning. Unlike traditional ME tasks, MEBench further incorporates spatial reasoning to create more challenging and realistic evaluation settings. We assess the performance of state-of-the-art vision-language models (VLMs) on this benchmark using novel evaluation metrics that capture key aspects of ME-based reasoning. To facilitate controlled experimentation, we also present a flexible and scalable data generation pipeline that supports the construction of diverse annotated scenes.

ShotAdapter: Text-to-Multi-Shot Video Generation with Diffusion Models

Current diffusion-based text-to-video methods are limited to producing short video clips of a single shot and lack the capability to generate multi-shot videos with discrete transitions where the same character performs distinct activities across the same or different backgrounds. To address this limitation we propose a framework that includes a dataset collection pipeline and architectural extensions to video diffusion models to enable text-to-multi-shot video generation. Our approach enables generation of multi-shot videos as a single video with full attention across all frames of all shots, ensuring character and background consistency, and allows users to control the number, duration, and content of shots through shot-specific conditioning. This is achieved by incorporating a transition token into the text-to-video model to control at which frames a new shot begins and a local attention masking strategy which controls the transition token's effect and allows shot-specific prompting. To obtain training data we propose a novel data collection pipeline to construct a multi-shot video dataset from existing single-shot video datasets. Extensive experiments demonstrate that fine-tuning a pre-trained text-to-video model for a few thousand iterations is enough for the model to subsequently be able to generate multi-shot videos with shot-specific control, outperforming the baselines. You can find more details in https://shotadapter.github.io/

SocialGesture: Delving into Multi-person Gesture Understanding

Previous research in human gesture recognition has largely overlooked multi-person interactions, which are crucial for understanding the social context of naturally occurring gestures. This limitation in existing datasets presents a significant challenge in aligning human gestures with other modalities like language and speech. To address this issue, we introduce SocialGesture, the first large-scale dataset specifically designed for multi-person gesture analysis. SocialGesture features a diverse range of natural scenarios and supports multiple gesture analysis tasks, including video-based recognition and temporal localization, providing a valuable resource for advancing the study of gesture during complex social interactions. Furthermore, we propose a novel visual question answering (VQA) task to benchmark vision language models'(VLMs) performance on social gesture understanding. Our findings highlight several limitations of current gesture recognition models, offering insights into future directions for improvement in this field. SocialGesture is available at huggingface.co/datasets/IrohXu/SocialGesture.

Learning Predictive Visuomotor Coordination

Understanding and predicting human visuomotor coordination is crucial for applications in robotics, human-computer interaction, and assistive technologies. This work introduces a forecasting-based task for visuomotor modeling, where the goal is to predict head pose, gaze, and upper-body motion from egocentric visual and kinematic observations. We propose a \textit{Visuomotor Coordination Representation} (VCR) that learns structured temporal dependencies across these multimodal signals. We extend a diffusion-based motion modeling framework that integrates egocentric vision and kinematic sequences, enabling temporally coherent and accurate visuomotor predictions. Our approach is evaluated on the large-scale EgoExo4D dataset, demonstrating strong generalization across diverse real-world activities. Our results highlight the importance of multimodal integration in understanding visuomotor coordination, contributing to research in visuomotor learning and human behavior modeling.

Towards Online Multi-Modal Social Interaction Understanding

Multimodal social interaction understanding (MMSI) is critical in human-robot interaction systems. In real-world scenarios, AI agents are required to provide real-time feedback. However, existing models often depend on both past and future contexts, which hinders them from applying to real-world problems. To bridge this gap, we propose an online MMSI setting, where the model must resolve MMSI tasks using only historical information, such as recorded dialogues and video streams. To address the challenges of missing the useful future context, we develop a novel framework, named Online-MMSI-VLM, that leverages two complementary strategies: multi-party conversation forecasting and social-aware visual prompting with multi-modal large language models. First, to enrich linguistic context, the multi-party conversation forecasting simulates potential future utterances in a coarse-to-fine manner, anticipating upcoming speaker turns and then generating fine-grained conversational details. Second, to effectively incorporate visual social cues like gaze and gesture, social-aware visual prompting highlights the social dynamics in video with bounding boxes and body keypoints for each person and frame. Extensive experiments on three tasks and two datasets demonstrate that our method achieves state-of-the-art performance and significantly outperforms baseline models, indicating its effectiveness on Online-MMSI. The code and pre-trained models will be publicly released at: https://github.com/Sampson-Lee/OnlineMMSI.