Search papers, labs, and topics across Lattice.
This paper introduces ChronoVision, a multimodal framework that enhances temporal reasoning in large language models by aligning visual logic with latent imagery through a novel Reconstructive Visual Head and an ROI Attention Locating module. The approach leverages supervised fine-tuning and reinforcement learning with a composite reward function to improve model performance on complex visual cognitive tasks. Experimental results show that ChronoVision achieves state-of-the-art accuracy on the newly proposed Vbvr-VQA dataset and the challenging IntPhys2 benchmark, indicating significant advancements in visual reasoning capabilities.
Achieving 74.8% accuracy on a new temporal reasoning benchmark, ChronoVision redefines how multimodal models can tackle complex visual tasks.
Multimodal large language models excel at passive perception but struggle with complex visual cognitive tasks requiring multi-step temporal reasoning. This degradation largely stems from the inherent ambiguity of language-based reasoning, which often fails to accurately articulate continuous visual transformations. To address this, we propose ChronoVision, a multimodal framework designed to align visual logic with latent imagery. During supervised fine-tuning, a Reconstructive Visual Head predicts the latent representation of the final transformed state, while an ROI Attention Locating module focuses the model on key visual evidence via semantic span queries. In post-training, we apply reinforcement learning with an implicit process grounding mechanism, guided by a composite reward function that evaluates outcome correctness, latent process alignment, and unsupervised visual focus. Furthermore, we introduce Vbvr-VQA, a novel dataset that evaluates temporal tracking by reformulating video reasoning into a strict image-ordering task. Experiments demonstrate that ChronoVision achieves state-of-the-art performance on Vbvr-VQA with 74.8% in-domain and 71.6% out-of-domain accuracy, alongside a strong 55.0% accuracy on IntPhys2, a highly challenging cross-domain benchmark.
