How to call the trained weights for testing

[zhaiyuyang]

It seems that the model saves the weights in the format net_xxx.pth during training. However, during testing, I made the following modifications according to your method：
self.model_t = Namespace() self.model_t.name = 'vit_small_patch16_224_dino' self.model_t.kwargs = dict(pretrained=False, checkpoint_path= 'model/pretrain/net_100.pth', pretrained_strict=False, strict=True, img_size=self.size, teachers=[3, 6, 9], neck=[12]) # self.model_t.kwargs = dict(pretrained=True, checkpoint_path='', strict=True, img_size=self.size, teachers=[3, 6, 9], neck=[12]) self.model_f = Namespace() self.model_f.name = 'fusion' self.model_f.kwargs = dict(pretrained=False, checkpoint_path='', strict=False, dim=384, mul=1) self.model_s = Namespace() self.model_s.name = 'de_vit_small_patch16_224_dino' self.model_s.kwargs = dict(pretrained=False, checkpoint_path='', strict=False, img_size=self.size, students=[3, 6, 9], depth=9) self.model = Namespace() self.model.name = 'vitad' self.model.kwargs = dict(pretrained=False, checkpoint_path='', strict=True, model_t=self.model_t, model_f=self.model_f, model_s=self.model_s)
But an error occurred during execution：
RuntimeError: Error(s) in loading state_dict for ViT_Encoder:`，"Clearly, the weight file does not match the model input."。
"Could you please advise me on how to resolve this? Thank you in advance for your response."

[zhaiyuyang]

Can someone come to help me?

[zhangzjn]

You can modify trainer.resume_dir to LOGDIR, or model.kwargs['checkpoint_path'] to LOGDIR/net_xxx.pth. I hope this helps you.