You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This is a question/request more than an actual issue. First of all, thanks very much for releasing the inference model. I have questions with respect to the difference between training and inference. From the deploy prototxt, the inference model includes VGG to compute parts, and VGGs for classification using each parts and the whole images. Fig 2 in the paper just has one conv layer in (b). Furthermore, the inference model takes the mask max pos as center and 96x96 crop for part image, while training uses equation (5) to compute the part feature. I will be very appreciated if you could explain the training process compared with sec 3.3 of the paper. Have the training prototxt released would be the definite answer.
The text was updated successfully, but these errors were encountered:
This is a question/request more than an actual issue. First of all, thanks very much for releasing the inference model. I have questions with respect to the difference between training and inference. From the deploy prototxt, the inference model includes VGG to compute parts, and VGGs for classification using each parts and the whole images. Fig 2 in the paper just has one conv layer in (b). Furthermore, the inference model takes the mask max pos as center and 96x96 crop for part image, while training uses equation (5) to compute the part feature. I will be very appreciated if you could explain the training process compared with sec 3.3 of the paper. Have the training prototxt released would be the definite answer.
The text was updated successfully, but these errors were encountered: