ValueError: operands could not be broadcast together with shapes (256,256,4) (3,) (256,256,4)

[abhishekkrthakur]

ValueError: operands could not be broadcast together with shapes (256,256,4) (3,) (256,256,4)

[abhishekkrthakur]

It seems like your image is RGBA. Can you convert it to RGB? or do you have to use RGBA? If latter, then try writing your own dataloader.
Can you provide data and code to reproduce the error?

[nvnvashisth]

Ok I converted everything to RGB. I have label from [0-9] with image size 256x256. But I come across this CUDA error. Another thing, I am trying to execute in Colab.

Regarding the code, it is exactly taken from here https://www.kaggle.com/abhishek/tez-faster-and-easier-training-for-leaf-detection ;)

Loaded pretrained weights for efficientnet-b4

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-29-de0853739e51> in <module>()
     11     epochs=10,
     12     callbacks=[es],
---> 13     fp16=True
     14 )
     15 model.save("model.bin")

6 frames
/usr/local/lib/python3.6/dist-packages/tez/model/model.py in fit(self, train_dataset, valid_dataset, train_sampler, valid_sampler, device, epochs, train_bs, valid_bs, n_jobs, callbacks, fp16)
    289             n_jobs=n_jobs,
    290             callbacks=callbacks,
--> 291             fp16=fp16,
    292         )
    293 

/usr/local/lib/python3.6/dist-packages/tez/model/model.py in _init_model(self, device, train_dataset, valid_dataset, train_sampler, valid_sampler, train_bs, valid_bs, n_jobs, callbacks, fp16)
     81 
     82         if next(self.parameters()).device != device:
---> 83             self.to(device)
     84 
     85         if self.train_loader is None:

/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py in to(self, *args, **kwargs)
    610             return t.to(device, dtype if t.is_floating_point() else None, non_blocking)
    611 
--> 612         return self._apply(convert)
    613 
    614     def register_backward_hook(

/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py in _apply(self, fn)
    357     def _apply(self, fn):
    358         for module in self.children():
--> 359             module._apply(fn)
    360 
    361         def compute_should_use_set_data(tensor, tensor_applied):

/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py in _apply(self, fn)
    357     def _apply(self, fn):
    358         for module in self.children():
--> 359             module._apply(fn)
    360 
    361         def compute_should_use_set_data(tensor, tensor_applied):

/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py in _apply(self, fn)
    379                 # `with torch.no_grad():`
    380                 with torch.no_grad():
--> 381                     param_applied = fn(param)
    382                 should_use_set_data = compute_should_use_set_data(param, param_applied)
    383                 if should_use_set_data:

/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py in convert(t)
    608             if convert_to_format is not None and t.dim() == 4:
    609                 return t.to(device, dtype if t.is_floating_point() else None, non_blocking, memory_format=convert_to_format)
--> 610             return t.to(device, dtype if t.is_floating_point() else None, non_blocking)
    611 
    612         return self._apply(convert)

RuntimeError: CUDA error: device-side assert triggered

[abhishekkrthakur]

Code provided in examples works quite well. This seems like some problem with the model. I cant say without having data and full code to reproduce the error :)

[abhishekkrthakur]

@nvnvashisth I just added a multi-class classification example (flower classification with 104 classes). It might be useful for you: https://github.com/abhishekkrthakur/tez/blob/main/examples/image_classification/flower_classification.py

Let me know if it still doesnt work.

[nvnvashisth]

Code provided in examples works quite well. This seems like some problem with the model. I cant say without having data and full code to reproduce the error :)

I have the code privately on your twitter (DM). That's the only way I could figure to reach you privately.

[nvnvashisth]

@nvnvashisth I just added a multi-class classification example (flower classification with 104 classes). It might be useful for you: https://github.com/abhishekkrthakur/tez/blob/main/examples/image_classification/flower_classification.py

Let me know if it still doesnt work.

I'll give it a try. Thanks

[nvnvashisth]

@abhishekkrthakur it is so weird. I didn't really change anything and it started working. No more cuda error. Thanks for quick support.

[abhishekkrthakur]

wow. maybe you updated torch?

[nvnvashisth]

Not really, I was running in colab, was using the default one.

[agg-shambhavi]

It seems like your image is RGBA. Can you convert it to RGB? or do you have to use RGBA? If latter, then try writing your own dataloader.

@abhishekkrthakur
Thank you! This helped me to solve the above error.