Error with training RIMES

[huynhtruc0309]

Thank you for your great work.

1. I am trying to train HWR on RIMES dataset by this command python train.py -c configs/cf_RIMESLines_hwr_cnnOnly_batchnorm_aug.json

But the error happens.

Begin training
Traceback (most recent call last):
  File "train.py", line 132, in <module>
    main(config, args.resume)
  File "train.py", line 78, in main
    trainer.train()
  File "handwriting_line_generation/base/base_trainer.py", line 219, in train
    result = self._train_iteration(self.iteration)
  File "handwriting_line_generation/trainer/hw_with_style_trainer.py", line 378, in _train_iteration
    pred, recon, losses = self.run(instance)
  File "handwriting_line_generation/trainer/hw_with_style_trainer.py", line 736, in run
    pred = self.model.hwr(image, style)
  File "lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "handwriting_line_generation/model/cnn_only_hwr.py", line 132, in forward
    conv = self.cnn(input)
  File "lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "lib/python3.8/site-packages/torch/nn/modules/container.py", line 117, in forward
    input = module(input)
  File "lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "lib/python3.8/site-packages/torch/nn/modules/conv.py", line 423, in forward
    return self._conv_forward(input, self.weight)
  File "lib/python3.8/site-packages/torch/nn/modules/conv.py", line 419, in _conv_forward
    return F.conv2d(input, weight, self.bias, self.stride,
RuntimeError: Input type (torch.cuda.LongTensor) and weight type (torch.cuda.FloatTensor) should be the same

Then I add input = input.type(torch.cuda.FloatTensor) to this function in cnn_only_hwr.py

def forward(self, input, style=None):
        if self.pad is not None:
            input=self.pad(input)
        # conv features
        input = input.type(torch.cuda.FloatTensor)
        conv = self.cnn(input)
        b, c, h, w = conv.size()
        # assert h == 1, "the height of conv must be 1"
        conv = conv.view(b, -1, w)
        conv = self.cnn1d(conv)
        output = conv.permute(2, 0, 1)  # [w, b, c]

        return output

Then this error happens:

Begin training
{'recogLoss': nan}
Traceback (most recent call last):
  File "train.py", line 132, in <module>
    main(config, args.resume)
  File "train.py", line 78, in main
    trainer.train()
  File "/home/ocr/truchlp/handwriting_line_generation/base/base_trainer.py", line 219, in train
    result = self._train_iteration(self.iteration)
  File "/home/ocr/truchlp/handwriting_line_generation/trainer/hw_with_style_trainer.py", line 399, in _train_iteration
    assert(loss==0 or (not torch.isnan(loss) and not torch.isinf(loss)))
AssertionError

2. I am also trying to train GAN on RIMES dataset by this command python train.py -c configs/cf_RIMESLinesslant_noMask_charSpecSingleAppend_GANMedMT_autoAEMoPrcp2tightNewCTCUseGen_balB_hCF0.75_sMG.json
   But it says:

Set valid_data_loader to None
Traceback (most recent call last):
  File "train.py", line 132, in <module>
    main(config, args.resume)
  File "train.py", line 62, in main
    trainer = trainerClass(model, loss, metrics,
  File "handwriting_line_generation/trainer/hw_with_style_trainer.py", line 115, in __init__
    self.text_data = TextData(config['trainer']['text_data'],config['data_loader']['char_file'],text_data_batch_size,max_len=text_data_max_len,words=text_words,characterBalance=characterBalance) if 'text_data' in config['trainer'] else None
  File "handwriting_line_generation/datasets/text_data.py", line 15, in __init__
    with open(textfile) as f:
FileNotFoundError: [Errno 2] No such file or directory: 'data/french_news.txt'

Thank you for helping me.

[herobd]

I made a fix with I think will help with the first error.
The french_news.txt is derived from this database: https://webhose.io/free-datasets/french-news-articles/
I've added the python script I used to compile it's jsons into the single txt file: data/compile_french_new.py. First cre
ate a file listing all the jsons (ls > jsons.txt) then run the script.

However, any French text corpus should do, it isn't that important as it just is providing text samples that aren't in the RIMES dataset.