two-stage training strategy
- COS3D: Collaborative Open-Vocabulary 3D Segmentation
- Efficient Rectified Flow for Image Fusion
- SceneDesigner: Controllable Multi-Object Image Generation with 9-DoF Pose Manipulation
- ShotBench: Expert-Level Cinematic Understanding in Vision-Language Models
- Target Speaker Extraction through Comparing Noisy Positive and Negative Audio Enrollments
- VADB: A Large-Scale Video Aesthetic Database with Professional and Multi-Dimensional Annotations