CVPR 2026 • 2026
We propose ReMoRa, a video MLLM that processes videos by operating directly on their compressed representations, using sparse RGB keyframes for appearance and a refined motion representation for temporal dynamics.
CVPR 2026 Findings • 2026
We propose NaiLIA, a multimodal retrieval method for nail design images that comprehensively aligns with dense intent descriptions and palette queries.
arXiv • 2025
We present the AIRoA MoMa Dataset, a large-scale hierarchical dataset designed to advance research in mobile manipulation within indoor environments.
IEEE RA-L • 2025
In this study we propose RelaX-Former, a method that leverages unlabeled positive labels and introduces a double relaxed contrastive learning approach to handle unlabeled positive and negative samples, improving the alignment between images and text.
IEEE RA-L • 2025
In this study we propose a novel training method that leverages both learning-based and n-gram based automatic evaluation metrics as rewards to generate free-form mobile manipulation instructions.