arXiv Computer Vision research summaries for March 10, 2024.

Today's Research Themes (AI-Generated):

• Introduction of VidProM, the first large-scale dataset for text-to-video prompts with 1.67 million unique entries, revolutionizing text-to-video diffusion model research.

• Proposing Temporally Coherent Action model, a novel video data replay technique for incremental action segmentation, outperforming baselines by up to 22% on the Breakfast dataset.

• UDE, a Universal Debiased Editing strategy for fair medical image classification, mitigates biases in Foundation Models' APIs, enhancing AI-driven medicine equity.

• Development of edge-based approach for textureless object recognition, showcasing RGB images enhanced with edge features outperform in accuracy in real-time applications.

• CLEAR, a unified network leveraging cross-transformers and a pre-trained language model, achieves state-of-the-art performance in person attribute recognition and retrieval.