search

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
708 stars 130 forks source link

freihand_train_data.json format #132

Open souldyf opened 1 month ago

souldyf commented 1 month ago

Hi! Thanks for your excellent work! I have some quetions on freihand_train_data.json format.In the training_mano.json annotation file provided with the Freihand dataset, the first 48 denote poses, 49-58 are shapes, and the last three are trans; in your freihand_train_data.json file, the mano_param is divided into pose, shape, and trans, where the shape part is the same as the shape part provided in the Freihand dataset, but the pose and trans parts are different, can you tell me what transformations are done?

mks0601 commented 1 month ago

I converted all world coordinates to the camera coordinates.

souldyf commented 1 month ago

Thank you for your answer! Could you give some codes or reference articles on how to convert this coordinate?