re-implementation of DVF

origin link: https://github.com/lxx1991/pytorch-voxel-flow

we re-implement the sync_bn using cupy >= 5.0.0, and provide a test script for frame prediction the frame interpolation is not test

running envs: cupy >= 5.0.0, pytorch >= 1.0.0

test prediction

1. download the UCF101-test: https://drive.google.com/file/d/1EDIn8gxHpApmVPzDts2zc45ixbnqAegp/view
2. modify dataset path of core/datasets/ucf_101_test_new.py for example: dataset_path = '/DATA/wangshen_data/UCF101/ucf101_extrap_ours'
3. download the pre-trained weight: https://drive.google.com/file/d/1FB-mpS4UokiLriDBNJSBmozMQRH0Qez1/view
4. modify model path of test_prediction.py for example: checkpoint_path = '/DATA/wangshen_data/UCF101/voxelflow_finetune_model_best.pth.tar'
5. run the prediction process: python test_prediction.py configs/voxel-flow.py

runing results:

Testing Results: PSNR 31.457 (31.4573) Loss 0.00856 31.457312964287905

add test_prediction_vimeo.py

test next frame prediction on vimeo-90k sr dataset.