search

hulianyuyy / CorrNet

Continuous Sign Language Recognition with Correlation Network (CVPR 2023)
85 stars 14 forks source link

Weigh Tensor fill with tensor of 0 #3

Closed OssamaChrifi closed 10 months ago

OssamaChrifi commented 1 year ago

Hello, i try to run your work in windows systeme but i had some error :

Here we see, with the debug, that the weight tensor is correctly initialized :

image

but when this commande is run in the seq_scripts function :

for batch_idx, data in enumerate(tqdm(loader))

the weigh tensor is put to 0 :

image

Systeme : Pythorch 2.0

i work with 1 GPU, when i run the resnet.py alone the wheigh is correctly initialized and the convolution work correctly.

when i try to entre in this line with the debug : for batch_idx, data in enumerate(tqdm(loader))

it send me in the file comm.py in /modules/sync_batchnorm in the class SyncMaster :

def getstate(self): return {'master_callback': self._master_callback}

hulianyuyy commented 1 year ago

Sorry, i can't give an exact answer. But i suppose that the initializations are skipped for these tensors if you enter this line with the debug : for batch_idx, data in enumerate(tqdm(loader)).