环境安装指南请参考mmflow.
本次实践用到的模型以及下载地址如下表：
| Models | Training Datasets | Log | Config | Download |
| FlowNetC+ft | Flying Chairs + Sintel | log | Config | Model |
|---|---|---|---|---|
| FlowNet2CSS-sd | Flying Chairs + FlyingThing3d subset + ChairsSDHom | log | Config | Model |
| PWC-Net+ | FlyingChairs + FlyingThing3d subset + Sintel + KITTI2015 + HD1K | log | config | model |
| RAFT | FlyingChairs + FlyingThing3d + Sintel + KITTI2015 + HD1K | log | Config | Model |
| GMA(+p) | FlyingChairs + FlyingThing3d + Sintel + KITTI2015 + HD1K | log | Config | Model |
将上表中各模型的Config与Model下载保存至 checkpoints 子目录中，在根目录下运行 demo/warp_optical_flow.py ，运行结果生成在 results 子目录中。