Real-time evaluation error

[terbed]

Hi!

I couldn't run the evaluation, there is something wrong with loading the model:

python evaluate_images.py --images /media/sztaki/pi/video5 --model pretrained/model_segmentation_realtime_skin_30.pth --model-type BiSeNetV2 --save

INFO:root:loading BiSeNetV2 from pretrained/model_segmentation_realtime_skin_30.pth
Traceback (most recent call last):
File "evaluate_images.py", line 58, in
model = load_model(models[args.model_type], torch.load(args.model))
File "/home/terbe/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/serialization.py", line 426, in load
return _load(f, map_location, pickle_module, **pickle_load_args)
File "/home/terbe/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/serialization.py", line 603, in _load
magic_number = pickle_module.load(f, **pickle_load_args)
_pickle.UnpicklingError: invalid load key, 'v'.

Is there any idea what could be the problem?

[WillBrennan]

It looks like its a pytorch version issue. I'd try creating the environment specified in this project and seeing if it happens again. Can you tell me what version of pytorch your using?

python -c 'import torch; print(torch.__version__)'

It should be 1.5.0 if you're using the enviroment?

The models in this repo are stored with git-lfs, the other option is that you didn't grab the models correctly. Their sizes might be zero when you look at them in file explorer?

To pull git-lfs files run,

git lfs pull

Or just download them directly from github. When I run it on mine it loads correctly,

(semantic_segmentation) will@will-workstation2:~/code/semantic_segmentation$ python evaluate_images.py --images /media/sztaki/pi/video5 --model pretrained/model_segmentation_realtime_skin_30.pth --model-type BiSeNetV2 --save
INFO:root:loading BiSeNetV2 from pretrained/model_segmentation_realtime_skin_30.pth
INFO:root:creating model with categories: ['skin']
INFO:root:evaluating images from /media/sztaki/pi/video5
Traceback (most recent call last):
  File "evaluate_images.py", line 72, in <module>
    for image_file in find_files(image_dir, ['.png', '.jpg', '.jpeg']):
  File "evaluate_images.py", line 31, in find_files
    assert dir_path.exists()
AssertionError

[terbed]

Yes, installing a new version of pytorch solved the issue! Thank you!