On my Linux mint 19.1 using an RTX 2070

When trying to recognize using the default installation:

(p3p) home@home-lnx:~/Desktop/programs/P2PaLA$ python P2PaLA.py --config config_ALAR_min_model_17_12_18.txt --prev_model ALAR_min_model_17_12_18.pth --prod_data ./images/
2019-01-21 13:42:19,280 - optparse - INFO - Reading configuration from config_ALAR_min_model_17_12_18.txt
2019-01-21 13:42:19,282 - P2PaLA - INFO - Working on prod inference...
2019-01-21 13:42:19,283 - P2PaLA - INFO - Results will be saved to ./work/results/prod
2019-01-21 13:42:19,599 - P2PaLA - INFO - Resumming from model ALAR_min_model_17_12_18.pth
/home/home/.conda/envs/p3p/lib/python3.6/site-packages/torch/cuda/__init__.py:95: UserWarning: 
    Found GPU0 GeForce RTX 2070 which requires CUDA_VERSION >= 9000 for
     optimal performance and fast startup time, but your PyTorch was compiled
     with CUDA_VERSION 8000. Please install the correct PyTorch binary
     using instructions from http://pytorch.org
    
  warnings.warn(incorrect_binary_warn % (d, name, 9000, CUDA_VERSION))

So I installed latest torch and torchvision:

(p3p) home@home-lnx:~/Desktop/programs/P2PaLA$ pip install --ignore-installed torch torchvision

Then ran recognition:

(p3p) home@home-lnx:~/Desktop/programs/P2PaLA$ python P2PaLA.py --config config_ALAR_min_model_17_12_18.txt --prev_model ALAR_min_model_17_12_18.pth --prod_data ./images/
/home/home/.conda/envs/p3p/lib/python3.6/site-packages/torch/nn/_reduction.py:49: UserWarning: size_average and reduce args will be deprecated, please use reduction='mean' instead.
  warnings.warn(warning.format(ret))
2019-01-21 13:58:31,771 - optparse - INFO - Reading configuration from config_ALAR_min_model_17_12_18.txt
2019-01-21 13:58:31,773 - P2PaLA - INFO - Working on prod inference...
2019-01-21 13:58:31,774 - P2PaLA - INFO - Results will be saved to ./work/results/prod
2019-01-21 13:58:32,125 - P2PaLA - INFO - Resumming from model ALAR_min_model_17_12_18.pth
2019-01-21 13:58:34,859 - P2PaLA - INFO - Preprocessing data from ./images/
P2PaLA.py:1195: UserWarning: volatile was removed and now has no effect. Use `with torch.no_grad():` instead.
  pr_x = Variable(sample["image"], volatile=True)
THCudaCheck FAIL file=/pytorch/aten/src/THC/THCGeneral.cpp line=405 error=11 : invalid argument
2019-01-21 13:58:35,463 - P2PaLA - INFO - Production stage done. total time taken: 0.604010820388794
2019-01-21 13:58:35,463 - P2PaLA - INFO - Average time per page: 0.604010820388794
2019-01-21 13:58:35,463 - P2PaLA - INFO - All Done...

Now the problem is when trying to train

(p3p) home@home-lnx:~/Desktop/programs/P2PaLA$ python P2PaLA.py --config config_BL_only.txt --tr_data ./data/train --te_data ./data/test --log_comment "_foo"
/home/home/.conda/envs/p3p/lib/python3.6/site-packages/torch/nn/_reduction.py:49: UserWarning: size_average and reduce args will be deprecated, please use reduction='mean' instead.
  warnings.warn(warning.format(ret))
2019-01-21 14:06:09,788 - optparse - INFO - Reading configuration from config_BL_only.txt
2019-01-21 14:06:09,789 - optparse - DEBUG - Creating output dir: ./work_BL_only
2019-01-21 14:06:09,790 - optparse - DEBUG - Creating checkpoints dir: ./work_BL_only/checkpoints
2019-01-21 14:06:09,790 - P2PaLA - INFO - Working on training stage...
2019-01-21 14:06:09,791 - P2PaLA - WARNING - tensorboardX is not installed, display logger set to OFF.
2019-01-21 14:06:09,791 - P2PaLA - INFO - Preprocessing data from ./data/train
/home/home/Desktop/programs/P2PaLA/nn_models/models.py:293: UserWarning: nn.init.uniform is now deprecated in favor of nn.init.uniform_.
  init.uniform(m.weight.data, 0.0, 0.02)
/home/home/Desktop/programs/P2PaLA/nn_models/models.py:298: UserWarning: nn.init.uniform is now deprecated in favor of nn.init.uniform_.
  init.uniform(m.weight.data, 1.0, 0.02)
THCudaCheck FAIL file=/pytorch/aten/src/THC/THCGeneral.cpp line=405 error=11 : invalid argument
Traceback (most recent call last):
  File "P2PaLA.py", line 1262, in <module>
    main()
  File "P2PaLA.py", line 606, in main
    epoch_lossD += d_loss.data[0]
IndexError: invalid index of a 0-dim tensor. Use tensor.item() to convert a 0-dim tensor to a Python number

Hi,

I want to extend this work to find the regions on different documents with different regions (other than defined in the regions list regions = ["$tip", "$par", "$not", "$nop", "$pag"]. Is this possible? If so, what change will be needed ? I am new to segmentation problems, please guide.

@lquirosd

Please advise on utility to draw the output baseline polygon on source image. I see an example on your online page(btw it doesn't work this time, skips the Process button). But I can't fond a way to look at it on my machine. If any built-in function please write the cli command: cmd data/source_img.jpg work/results_prod/page/source_img.xml
-> to produce drawn countor on source image. In my run there are zero-size jpg files in work/results/prod
Probably it's a sort of error so it's interesting to draw xml lines over image(s).

Hello. I have ran P2Pala succesfully against typewritten and print with the default model, but I am not getting very good results when running against handwriting. Is it possible to train a model to work against handwriting, and if so, what kind of ground truth, and how much is required?

@lquirosd

Currently, I am trying to train p2pala to recognize the "TextLine" regions not baselines. How exactly can I do that, how can I select the default TextLine region itself

An example page-xml is attached along with my training config txt file. sample.zip

Waiting for your reply

I've used P2PaLA to train a Document Layout Analysis model for zone segmentation with PRImA Layout Analysis Dataset. PRImA dataset use PAGE XML with 2010 schema version. When model load data from corpus, I've got:

No region type defined for r1 at 0000001096 Element type "Node" undefined on color dic, set to default=175

This message have happened for all <TextRegion> in the XML file. After training phase have done, I see the result/test and nothing was predicted. I don't know what it mean. Please help me. Thanks.

the error is as shown below: 2020-04-02 09:33:10,254 - optparse - INFO - Reading configuration from config_ALAR_min_model_17_12_18_inference.txt 2020-04-02 09:33:10,263 - P2PaLA - INFO - Working on prod inference... 2020-04-02 09:33:10,268 - P2PaLA - INFO - Results will be saved to ./work/results/prod 2020-04-02 09:33:10,844 - P2PaLA - INFO - Resumming from model ALAR_min_model_17_12_18.pth 2020-04-02 09:33:11,408 - P2PaLA - INFO - Preprocessing data from ./images Premature end of JPEG file Traceback (most recent call last): File "P2PaLA.py", line 1268, in main() File "P2PaLA.py", line 1250, in main out_folder=res_path, File "/content/gdrive/My Drive/P2PaLA-master/data/imgprocess.py", line 187, in gen_page os.path.realpath(self.img_data[img_id]), os.path.join(out_folder, img_name) File "/content/gdrive/My Drive/P2PaLA-master/data/imgprocess.py", line 503, in symlink_force raise e File "/content/gdrive/My Drive/P2PaLA-master/data/imgprocess.py", line 497, in symlink_force os.symlink(target, link_name) OSError: [Errno 95] Operation not supported: '/content/gdrive/My Drive/P2PaLA-master/images/5.jpeg' -> './work/results/prod/5.jpeg'

The requirements.txt currently lists opencv-python, which drags in libraries for windowing systems like X11 – not ideal for headless servers. The Python bindings for OpenCV do offer other build variants like -contrib and -headless though (which are on PyPI for many platforms, and can also be built by ENABLE_HEADLESS=1 python setup.py bdist_wheel).

Therefore I suggest switching to opencv-python-headless instead.

Hello, I played a little with the provided model with the weights ALAR_min_model_17_12_18.pth As to the word min in the name I wonder ther are other models. Do you plan to publish it?

I've thouigh out some tricks to improve the accuracy of coverage the htr text regions but it's still rough. In many cases important parts of letters are cropped out. I don't have neither hardware nor labeled datasets for the trainig. Could you share a more powerful model?

To adapt the script to extract information/coordinates about page, from XML's formats most knowledge in the industry, like YOLO and PASCAL/VOC.

Observation: I could help with this task.

@JoseRPrietoF What is different in your version of P2PaLA?

Modified version to do table segmentation and act separation on Passau and Chancery corpus.

• what do you mean by act separation?
• also, when you say table segmentation, do you mean it detects baseline, and then extracts text?

Bumps opencv-python from 3.4.5.20 to 4.2.0.32.

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.