Hello everyone! Good work!
I'm trying to develop a semi supervised semantic segmentation pipeline using a custom dataset. I was able to train the teacher model using labelled data only (pretraining) with the SWIN tiny backbone and saved those weights (TEACHER_CKPT). I have downloaded and converted the SWIN pretrained weights as mentioned in Mask2Former repo for this(swin-to-d2). However, when i train the semi supervised model i encounter this error "Invalid magic number". But i do not encounter this error with the RESNET backbones which also uses pretrained weights as .pkl.
Can someone help me out with this issue? How do i solve this?
` Traceback (most recent call last):
783 File "train_net.py", line 435, in <module>
784 launch(
785 File "/share/bharath/detectron2/detectron2/engine/launch.py", line 84, in launch
786 main_func(*args)
787 File "train_net.py", line 428, in main
788 trainer.resume_or_load(resume=args.resume)
789 File "/share/bharath/Mask2Former_SSL/modules/defaults.py", line 483, in resume_or_load
790 self.modified_checkpointer(resume)
791 File "/share/bharath/Mask2Former_SSL/modules/defaults.py", line 451, in modified_checkpointer
792 which = self.parse_ckpt(self.cfg.MODEL.WEIGHTS)
793 File "/share/bharath/Mask2Former_SSL/modules/defaults.py", line 427, in parse_ckpt
794 ckpt = torch.load(ckpt_dir, map_location='cpu')['model']
795 File "/share/bharath/miniconda3/envs/mask2former/lib/python3.8/site-packages/torch/serialization.py", line 608, in load
796 return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
797 File "/share/bharath/miniconda3/envs/mask2former/lib/python3.8/site-packages/torch/serialization.py", line 779, in _legacy_load
798 raise RuntimeError("Invalid magic number; corrupt file?")
799 RuntimeError: Invalid magic number; corrupt file?`
Hello everyone! Good work! I'm trying to develop a semi supervised semantic segmentation pipeline using a custom dataset. I was able to train the teacher model using labelled data only (pretraining) with the SWIN tiny backbone and saved those weights (TEACHER_CKPT). I have downloaded and converted the SWIN pretrained weights as mentioned in Mask2Former repo for this(swin-to-d2). However, when i train the semi supervised model i encounter this error "Invalid magic number". But i do not encounter this error with the RESNET backbones which also uses pretrained weights as .pkl. Can someone help me out with this issue? How do i solve this?