search

facebookresearch / frankmocap

A Strong and Easy-to-use Single View 3D Hand+Body Pose Estimator
Other
2.14k stars 374 forks source link

Implementing the algorithm (training) #221

Open tamarloeub opened 1 year ago

tamarloeub commented 1 year ago

Hi, I'm trying to implement the algorithm from the paper and encountering some issues. I'm training on the FreiHand dataset for now. I normalize the images before inserting them to the model. I implemented the loss as described in the paper (and projected the 3D joints to 2D), but the output looks like garbage. Maybe I'm missing some pre-processing steps on the input images? Or maybe you have some insights you can provide beyond what's described in the paper? Thanks in advance

penincillin commented 1 year ago

@tamarloeub It's hard to say the problem given so less information. I would suggest you refer to some similar repos with train code released, such as HandMesh.