mks0601 / I2L-MeshNet_RELEASE

Official PyTorch implementation of "I2L-MeshNet: Image-to-Lixel Prediction Network for Accurate 3D Human Pose and Mesh Estimation from a Single RGB Image", ECCV 2020
MIT License
720 stars 128 forks source link

training the rootnet with Freihand Dataset. #122

Open tyq996 opened 1 year ago

tyq996 commented 1 year ago

Hello, I have got your amazing work, but the test result of the root points on Freihand you give in your 'bbox_root_freihand_output.json' is not as the same as I train the rootnet using the FreiHand Dataset. Can you provide me the rootnet train code on FreiHand dataset? Thanks a lot !!!!!!

mks0601 commented 1 year ago

Sorry I do not saved the code. It is actually not very hard to extend RootNet to hand case. Pleaes note that you should adjust physical size of the body (original. maybe 2 meter) to the hand (about 15 cm).

tyq996 commented 1 year ago

Thanks a lot, but the same code I run under my conda environment gets 10.30cm, but your best result is 5.96cm, it confuses me a lot.

tyq996 commented 1 year ago

The focal is meter, milimeter or centimeter?