arXiv Computer Vision research summaries for March 28, 2024.

Today's Research Themes (AI-Generated):

• CLAP4CLIP introduces Continual LeArning with Probabilistic finetuning for Vision-Language Models, enhancing reliability and reducing forgetting in continual learning tasks.

• QNCD presents Quantization Noise Correction Scheme for Diffusion Models to improve image synthesis quality in low-bit settings.

• MoDiTalker proposes a novel motion-disentangled diffusion model for high-quality talking head generation from audio.

• Uncertainty-Aware Deep Video Compression uses ensembles to better capture predictive uncertainty and achieve significant bit rate savings.

• Dyco leverages delta pose sequence representation for improved 3D human modeling, capturing appearance changes caused by motion inertia.