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
In the training phase the number of predicted RoIs + scores is determined by hyperparameters like POST_NMS_ROIS_TRAINING. Caffe inits layers to (1,21) or (1,84) and then they are resized to match the number of predictions, e.g. (1000, 21) and (1000,84). This is done in custom layers, and I don't have a problem with that. At the same time, in order to get the loss, layers like cls_score, bbox_pred, fc6/7, are also resized to match the size (while keeping the number of weights the same). I don't see how Caffe does it out of the box, but I can't find the wrapper for that. Could anyone help?
The text was updated successfully, but these errors were encountered:
In the training phase the number of predicted RoIs + scores is determined by hyperparameters like
POST_NMS_ROIS_TRAINING. Caffe inits layers to (1,21) or (1,84) and then they are resized to match the number of predictions, e.g.(1000, 21)and(1000,84). This is done in custom layers, and I don't have a problem with that. At the same time, in order to get the loss, layers likecls_score, bbox_pred, fc6/7, are also resized to match the size (while keeping the number of weights the same). I don't see how Caffe does it out of the box, but I can't find the wrapper for that. Could anyone help?The text was updated successfully, but these errors were encountered: