Skip to content
This repository has been archived by the owner on Nov 16, 2023. It is now read-only.

Latest commit

 

History

History

weights

README.md

Pre-trained Models

Here, we provide three pre-trained models,

  • 771Sep25-0735_500000.params, the pre-trained MaskFlownet-S model, which has been trained on Flying Chairs and Flying Things3D, and it should be evaluated on Sintel train + val (corresponding to one in step 2).

  • dbbSep30-1206_1000000.params, the pre-trained MaskFlownet-S model, which has been trained on Flying Chairs, Flying Things3D, and Sintel train, and it should be evaluated on Sintel val (corresponding to one in step 3).

  • 5adNov03-0005_1000000.params, the pre-trained MaskFlownet model, which has been trained on Flying Chairs, Flying Things3D, and Sintel train, and it should be evaluated on Sintel val (corresponding to one in step 6).

Evaluation

Network Checkpoint Sintel train + val Sintel val KITTI 2012 KITTI 2015
MaskFlownet-S abbSep15-1037_500000 2.33, 3.72 2.93, 5.35 4.69, 0.20 11.88, 0.29
MaskFlownet-S dbbSep30-1206_1000000 - 2.70, 4.07 3.25, 0.11 9.14, 0.18
MaskFlownet 5adNov03-0005_1000000 - 2.52, 3.83 2.85, 0.10 8.15, 0.17

for Sintel, the values are AEPE (clean), AEPE (final); for KITTI, the values are AEPE, FI-all.

Inferring

For example,

  • to do validation for MaskFlownet-S on checkpoint 771Sep25-0735_500000.params, run python main.py MaskFlownet_S.yaml -g 0 -c 771Sep25 --valid (the output will be under ./logs/val/).

  • to do prediction for MaskFlownet on checkpoint 5adNov03-0005_1000000.params, run python main.py MaskFlownet.yaml -g 0 -c 5adNov03 --predict (the output will be under ./flows/).