I don't have a json model about lipreading_using_temporal_convolutional_networks HOT 9 CLOSED

Hi, kategia,

We save the configuration in a format of json. You could acess the configuration (.json) at the folder of configs.

If you would like to evaluate the model, you could specify the pretrained mode (.pth.tar) to --model-path.

from lipreading_using_temporal_convolutional_networks.

Thank you for your response! I now get raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

from lipreading_using_temporal_convolutional_networks.

Hi, there

This thread (30664) may help

from lipreading_using_temporal_convolutional_networks.

Hello again,
I just removed the .cuda() part from extract_feats=args.extract_feats).cuda() since for now I run it locally thank you for the thread though. I get this error
line 215, in main
model, optimizer, epoch_idx, ckpt_dict = load_model(args.model_path, model, optimizer)
optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
KeyError: 'optimizer_state_dict'

I haven't modyfied the optimizer parameters here they are
parser.add_argument('--optimizer',type=str, default='adamw', choices = ['adam','sgd','adamw'])
Any idea what could fix this? Thank you

from lipreading_using_temporal_convolutional_networks.

Hi, kategia,

Could you tell me which checkpoint file did you use?

from lipreading_using_temporal_convolutional_networks.

Hello,
sure its the resnet18_mstcn(adamw_s3) model -->lrw_resnet18_mstcn_adamw_s3.pth.tar

from lipreading_using_temporal_convolutional_networks.

Hi,

The checkpoint only includes the weight of model. That's why no keys in optimiser.

If you are testing the model or initialise from our model, please leave '--init-epoch' to 0.

https://github.com/mpc001/Lipreading_using_Temporal_Convolutional_Networks/blob/master/main.py#L215-L219 is for loading model to resume training. The previous checkpoints, which includes the keys in optimiser, could be accessed in the folder of train_logs

from lipreading_using_temporal_convolutional_networks.

Yep my bad you are totally right.
One last thing,
when I try to train the model I get the following error
File "C:/lipreading/Lipreading_using_Temporal_Convolutional_Networks/main.py", line 240, in main
model = train(model, dset_loaders['train'], criterion, epoch, optimizer, logger)
File "C:/Lipreading_using_Temporal_Convolutional_Networks/main.py", line 135, in train
for batch_idx, (input, lengths, labels) in enumerate(dset_loader):
...............
assert os.path.isfile( filepath ), "Error when trying to read txt file, path does not exist: {}".format(filepath)
AssertionError: Error when trying to read txt file, path does not exist: C:\Lipreading_using_Temporal_Convolutional_Networks\datasets\visual_data\ABOUT\train\ABOUT_00849.txt

In this folder I have the ABOUT folder with the processed images(npz) format and no txt file. Where does the error come from?Thank youuu

from lipreading_using_temporal_convolutional_networks.

Hi,

The annonation file (.txt) is in the original LRW-BBC dataset. Also, please note that the splitting symbol for file path is different between windows and unix.

from lipreading_using_temporal_convolutional_networks.