类库
› VideoMaMa
cvlab-kaist/VideoMaMa
VideoMaMa是一个利用生成先验的掩码引导视频抠图框架，基于CVPR 2026研究实现。通过结合视频生成先验知识，能在多种视频场景下提供稳定且精细的抠图效果，适用于高质量视频前景分离任务。
技术栈
查看全部依赖 (17)
依赖
NumPy
unknown
Pillow
unknown
Pydantic
unknown
accelerate
unknown
diffusers
unknown
einops
unknown
gradio
unknown
imageio
unknown
imageio-ffmpeg
unknown
opencv-contrib-python
unknown
opencv-python
unknown
safetensors
unknown
scipy
unknown
torch
unknown
torchvision
unknown
tqdm
unknown
transformers
unknown
