【论文复现】摩尔线程版AnimateAnyone

code：MooreThreads/Moore-AnimateAnyone (github.com)

paper：Animate Anyone (humanaigc.github.io)

1. Inference

Inference没啥难度，按照readme来就行，就是分开下载权重比较慢，注意权重文件严格按照文档目录树来放。

python -m scripts.pose2vid --config ./configs/prompts/animation.yaml -W 512 -H 784 -L 64

默认的参数我cuda.OutOfMemoryError了（服务器12G显存），调小为-W 256 -H 392 -L 32就可以了，但是测试出来的demo效果非常差，视频放不上来

2. Training

Data Preparation

用MEAD数据集试一下

Extract keypoints from raw videos:

python tools/extract_dwpose_from_vid.py --video_root /path/to/your/video_dir

Extract the meta info of dataset:

python tools/extract_meta_info.py --root_path /path/to/your/video_dir --dataset_name anyone 

Stage1

Stage2