Closed lyhnsn closed 1 month ago
The bug you find is interesting, because i do not use the ARTrackSeq's forward function, in another way, in seq_trainer, we use the computer_sequence_loss instead of forward (or call function). Maybe i leave this just because i transfer the sequence training code from one-stage training. Moreover, thanks for your reply, i will update it when i finish the paper recently.
Thank you for your excellent work.
There may be a bug, in "lib/train/actors/artrack_seq.py", line 232,
loss, status = self.compute_losses(out_dict, data)
but it should be "self.compute_sequence_losses()"? As there is no
self.compute_losses
in this file but in "lib/train/actors/artrack.py".Furthermore, the input of these methods are sightly different.
Looking for you reply.