In this paper, we present the solution developed by our team, XInsight Lab, which achieved first place in Track 3 of the 4th EI-MIGA-IJCAI Challenge with a test accuracy of 0. 76923.
Summary generated by The Flow from the publisher's feed. The full article lives at Hugging Face Trending Papers.
arXiv:2604. 15280v2 Announce Type: replace-cross Abstract: Understanding emotions is a fundamental ability for intelligent systems to be able to interact with humans.
arXiv:2607. 12774v1 Announce Type: cross Abstract: This article presents our results for the 11th Affective Behavior Analysis in-the-Wild (ABAW) competition.
arXiv:2606. 06903v1 Announce Type: cross Abstract: Human image animation aims to generate a video from a static reference image, guided by pose information extracted from a driving video.
arXiv:2607. 00296v1 Announce Type: cross Abstract: Human motion forecasting in unconstrained real-world videos remains challenging due to the ambiguity of future behaviors and the presence of noisy multimodal observations.
arXiv:2603. 09731v3 Announce Type: replace-cross Abstract: Multimodal large language models (MLLMs) are increasingly considered as a foundation for embodied agents, yet it remains unclear whether they can reliably reason about the long-term physical consequences of actions from an egocentric viewpoint.
arXiv:2606. 28083v1 Announce Type: cross Abstract: Micro-expression recognition is challenging due to subtle and short-lived facial muscle movements.