@watersink After downloading uw3-framed-lines-degraded-000, I extracted the images to a folder called uw3. Then I ran the training command train_test.py and got an error message:

home@home-lnx:~/ocrsegment$ python3 train_test.py 
Traceback (most recent call last):
  File "train_test.py", line 254, in <module>
    train()
  File "train_test.py", line 194, in train
    dataset,one_epoch_num = get_tf_dataset(dataset_text_file="./uw3/label.txt",batch_size=batch_size,channels=batch_channel)
  File "train_test.py", line 63, in get_tf_dataset
    filenames, labels,one_epoch_num = read_labeled_image_list(dataset_text_file)
  File "train_test.py", line 53, in read_labeled_image_list
    with open(dataset_text_file,"r",encoding="utf-8") as f_l:
FileNotFoundError: [Errno 2] No such file or directory: './uw3/label.txt'

@watersink Thank you for your hard work

I have successfully trained a new model using train_test.py. ocrseg.ckpt-200.zip

But when using my model for segmentation, an error message appears:

home@home-lnx:~/ocrsegment$ python3 segmentation.py 
Traceback (most recent call last):
  File "segmentation.py", line 444, in <module>
    lines = seg.extract_textlines(image)
  File "segmentation.py", line 425, in extract_textlines
    if len(image.shape)!=2:
AttributeError: 'NoneType' object has no attribute 'shape'

@watersink I tested training on my other system with 30GB ram and Quadro M4000 graphics card and it worked there. Thanks!

Originally posted by @ghost in https://github.com/watersink/ocrsegment/issues/1#issuecomment-422783272

I tried it to train using GTX970M 16GB RAM and it failed. I tried Google Colab and it failed too. I also tried using half of the dataset, setting per_process_gpu_memory_fraction=0.7 but it did not work. There is no other process using GPU. I am looking at the code trying to figure out is there something defining the input shape but it seems ok. Have you ever encountered the same problem?

Thanks in advance.