Closed Dawn-LX closed 1 year ago
Hi @Dawn-LX,
This seems to be resulted from the same issue in #77. If the classifier weights have 92 classes instead of 81, it suggests you are using the original DETR implementation, not the custom one attached in the submodule.
Simply navigate to the detr
directory and checkout all changes until git status
shows no more modifications.
Fred.
Hi @Dawn-LX,
This seems to be resulted from the same issue in #77. If the classifier weights have 92 classes instead of 81, it suggests you are using the original DETR implementation, not the custom one attached in the submodule.
Simply navigate to the
detr
directory and checkout all changes untilgit status
shows no more modifications.Fred.
Than you for your quick reply! I will further discuss this in issue #77 .
hello, 1). when I try the random init model, it runs into list index out of range for this line
target_cls_idx = [self.object_class_to_target_class[obj.item()]
2). Then I tried to used the pre-trained UPT's checkpoint, but the state_dict mis-match.
My args' configure is correct. the default value of args.dataset is hicodet.
So I wander whether it is a bug or my problem ?
For 1)
For 2):