Testing.md

Pre-trained model

Download pre-trained model, which is the BlurHandNet trained on BlurHand.

Then, locate pretrained_BlurHandNet_BH in experiments/.

The testing command is below:

CUDA_VISIBLE_DEVICES=0 python src/test.py -opt options/test/pretrained_BlurHandNet_BH.yml

In default, it calculates MPJPE on three sequential hands, and MPVPE on current (middle) hand.

	MPJPE (past / current / future)	MPVPE (current)
BlurHandNet	18.04 / 16.78 / 18.18 mm	15.31 mm

Note that the value is slightly different from the original paper (18.08 / 16.80 / 18.21, 15.30 in paper), due to reproduction.

The test logs will be saved in experiments/exp_name/results.

(Optional) If you want to visualize mesh sequences into video, set the visualize_video option as True. (But it takes a long time.)

Your own trained model

If you trained your own BlurHandNet following training instruction, please run the below command to test your model.

CUDA_VISIBLE_DEVICES=0 python src/test.py -opt options/test/BlurHandNet_BH.yml