error about ccl HOT 10 CLOSED

Hi,

The problem is caused by the ucf101_json.py, which is updated now.
Could you try with the new ucf101_json.py file? The problem should be resolved now.

Thanks.

from ccl.

Hi,

Thank you it fixed the previous issue. but now i get the following error:

No such file or directory: 'results/ucf51_ccl_ai/val.json

from ccl.

and this error during training:

File "main.py", line 569, in
main_worker(-1, opt)
File "main.py", line 516, in main_worker
train_ai_epoch(epoch=i, data_loader=train_loader, model=model, image_model=image_model,
File "/home/abhishek/abhi_scr/CCL/training.py", line 409, in train_ai_epoch
loss_vm = criterion_ct_iv(features, features_iv, targets) + criterion_ct_iv(features_iv, features, targets)
File "/home/abhishek/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/abhishek/abhi_scr/CCL/loss/nce_loss.py", line 34, in forward
log_pos_softmax = - torch.log(pred_softmax + EPISILON) * (1 - self_mask).float()
File "/home/abhishek/.local/lib/python3.8/site-packages/torch/tensor.py", line 396, in rsub
return _C._VariableFunctions.rsub(self, other)
RuntimeError: Subtraction, the - operator, with a bool tensor is not supported. If you are trying to invert a mask, use the ~ or logical_not() operator instead.

from ccl.

Hi,
Have you trained and tested the model successfully?
After training, there will be three log files in the folder: train.log train_batch.log val.log.
After testing, there will be a json file: val.json.

You could probably check if these files are ready to see whether the training or testing is missing. It is likely that the training or testing has not been done that caused missing the val.json file.

from ccl.

For the training error, it is likely to be caused by pytorch version. which version is used? the default version is 1.1.0

from ccl.

and this error during training:

File "main.py", line 569, in
main_worker(-1, opt)
File "main.py", line 516, in main_worker
train_ai_epoch(epoch=i, data_loader=train_loader, model=model, image_model=image_model,
File "/home/abhishek/abhi_scr/CCL/training.py", line 409, in train_ai_epoch
loss_vm = criterion_ct_iv(features, features_iv, targets) + criterion_ct_iv(features_iv, features, targets)
File "/home/abhishek/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/abhishek/abhi_scr/CCL/loss/nce_loss.py", line 34, in forward
log_pos_softmax = - torch.log(pred_softmax + EPISILON) * (1 - self_mask).float()
File "/home/abhishek/.local/lib/python3.8/site-packages/torch/tensor.py", line 396, in rsub
return _C._VariableFunctions.rsub(self, other)
RuntimeError: Subtraction, the - operator, with a bool tensor is not supported. If you are trying to invert a mask, use the ~ or logical_not() operator instead.

the nce_loss.py file is updated, which could probably allow working in a newer PyTorch version.

from ccl.

Many Thanks. the problem with pytorch version still persist but I should be able to fix it as we know it is becuase of pytorch version.

from ccl.

Training code runs after downgrading the pytorch to 1.1.0. thanks

from ccl.

and this error during training:

File "main.py", line 569, in
main_worker(-1, opt)
File "main.py", line 516, in main_worker
train_ai_epoch(epoch=i, data_loader=train_loader, model=model, image_model=image_model,
File "/home/abhishek/abhi_scr/CCL/training.py", line 409, in train_ai_epoch
loss_vm = criterion_ct_iv(features, features_iv, targets) + criterion_ct_iv(features_iv, features, targets)
File "/home/abhishek/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/abhishek/abhi_scr/CCL/loss/nce_loss.py", line 34, in forward
log_pos_softmax = - torch.log(pred_softmax + EPISILON) * (1 - self_mask).float()
File "/home/abhishek/.local/lib/python3.8/site-packages/torch/tensor.py", line 396, in rsub
return _C._VariableFunctions.rsub(self, other)
RuntimeError: Subtraction, the - operator, with a bool tensor is not supported. If you are trying to invert a mask, use the ~ or logical_not() operator instead.

Hi.
The error stems from the compatibility of boolean operation for each pytorch version.
For the newer version of Pytorch, we can simply change the lines of nce_loss.py code as follow to run.

Line 34
log_pos_softmax = - torch.log(pred_softmax + EPISILON) * (1 - self_mask.float())
to
log_pos_softmax = - torch.log(pred_softmax + EPISILON) * ((~self_mask).float())

and

Line 36
log_softmax = log_pos_softmax.sum(1) / (1 - self_mask).sum(1).float() + log_neg_softmax.sum(1) / self_mask.sum(1).float()
to
log_softmax = log_pos_softmax.sum(1) / (~self_mask).sum(1).float() + log_neg_softmax.sum(1) / self_mask.sum(1).float()

Please correct me if anything wrong :)

from ccl.

Thanks for pointing it out. It looks correct to me.

from ccl.