question about pseudo-GT files

[canghaiyunfan]

I find your offered pseudo-GT file (MSCOCO_train_MANO_NeuralAnnot.json) contain pose, shape, trans for mano, focal ,princpt for camera. I want to know how to get trans, focal, princpt.

[mks0601]

Assuming a fixed camera intrinsnic (focal length: (5000,5000), princpt:(128,128)), I got smpl parameters with perspective camera model. Hence, I got smpl_pose, smpl_shape, and trans. Then, I multiplied an inverse affine transformation to camera intrinsic matrix K to transform cropped single person space to original image space (before crop and resize).

[canghaiyunfan]

Thanks for your reply! I have some question about trans in mano layer, Is it same as camera para that model predict?

[mks0601]

Could you clarify what 'camera para` means? Also, 'model' means a regressor? or MANO model?

[canghaiyunfan]

'camera para` is regressor predict,such as ‘cam_trans’ in this code. 'model' means a regressor.

I want to know how to get trans in mano layer when you generate pseudo-GT file.

[mks0601]

For the right hand, trans is exactly same with cam_trans of that code. For the left hand, trans is torch.stack((-cam_trans[:,0], cam_trans[:,1], cam_trans[:,2]),1) because cam_trans in that code is for the right hand.

[canghaiyunfan]

Got it, Thanks very much.