CUDA out of memory while running test.py

[sindhurk]

While running test.py with the following code: python test.py --img pubtabnet/val/PMC524480_006_00.png --model BEST_checkpoint_table.pth.tar --word_map_structure output/WORDMAP_STRUCTURE.json --word_map_cell output/WORDMAP_CELL.json I get the following error. I tried in colab, with 32G memory and get a similar error. Here is the error msg:

test.py:18: DeprecationWarning:     `imread` is deprecated!
    `imread` is deprecated in SciPy 1.0.0, and will be removed in 1.2.0.
    Use ``imageio.imread`` instead.
  img = imread(image_path)
test.py:22: DeprecationWarning:     `imresize` is deprecated!
    `imresize` is deprecated in SciPy 1.0.0, and will be removed in 1.2.0.
    Use ``skimage.transform.resize`` instead.
  img = imresize(img, (height_image, width_image))
Traceback (most recent call last):
  File "test.py", line 363, in <module>
    seq, alphas, hidden_states = structure_image_beam_search(encoder_out, decoder_structure, word_map_structure, beam_size=args.beam_size_structure)
  File "test.py", line 90, in structure_image_beam_search
    awe, alpha = decoder.attention(encoder_out, h)
  File "/home/sindhu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/sindhu/projects/tab2html/table-recognition/code/image_caption/models.py", line 88, in forward
    att1 = self.encoder_att(encoder_out)  # (batch_size, num_pixels, attention_dim)
  File "/home/sindhu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/sindhu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/modules/linear.py", line 93, in forward
    return F.linear(input, self.weight, self.bias)
  File "/home/sindhu/anaconda3/envs/py37/lib/python3.7/site-packages/torch/nn/functional.py", line 1692, in linear
    output = input.matmul(weight.t())
RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 3.82 GiB total capacity; 2.64 GiB already allocated; 27.12 MiB free; 2.65 GiB reserved in total by PyTorch)

[kewin1807]

You can show config GPU what you used. i tested in GPU which VRAM is 11GB and it worked !!

[sindhurk]

Well, my vram is 4gb. Also, what is the accuracy you are getting with your model and how many epochs did you train?

[kewin1807]

I have just trained 10 epochs, after train my TED SCORE is 0.62, it is very different from papers. This repo is my base implementation for paper, you can config and edit to get higher score.

[sindhurk]

Can you pls share your trained weights file. My mail id is sindhuraman610@gmail.com

[sindhuattaiger]

replace model.eval() with torch.no_grad() and it works on my machine with 4gb vram