How to dewarp deskewed image (on page level) using ocrd-anybaseocr-dewarp

[Aysoltan]

ocrd-anybaseocr-dewarp with the command:

ocrd-anybaseocr-dewarp -I OCR-D-DESKEW-PAGE -O OCR-D-DEWARP-PAGE -p '{\"pix2pixHD\":\"../ocr_all/pix2pixHD/\",\"latest_net_G.pth\":\"../ocr_all/pix2pixHD/models\"}'

resolves the following error:
ValueError: Error parsing '{\"pix2pixHD\":\"../ocr_all/pix2pixHD/\",\"latest_net_G.pth\":\"../ocr_all/pix2pixHD/models\"}': Expecting property name enclosed in double quotes: line 1 column 2 (char 1)

Any idea how to use ocrd-anybaseocr-dewarp properly?

[kba]

This is an issue witht he JSON syntax. There is a simpler variant of the -p parameter: -P (capital P). This allows you to specify individual parameter key-value pairs without the JSON syntax:

ocrd-anybaseocr-dewarp -I OCR-D-DESKEW-PAGE -O OCR-D-DEWARP-PAGE -P model_path /path/to/latest_net_G.pth'

But I would recommend that you download the model with ocrd resmgr which takes care of placing the model in the right location, so you don't have to specify the model at all:

ocrd resmgr download ocrd-anybaseocr-dewarp '*'
# once that is finished, no need for an explicit parameter model_path because the default should work now
ocrd-anybaseocr-dewarp -I OCR-D-DESKEW-PAGE -O OCR-D-DEWARP-PAGE

Where did you get the parameters from? They are outdated, if we have this in our documentation still, I would like to update that.

[Aysoltan]

The parameters are from: [https://ocr-d.de/en/workflows] (look there at Step 6: Dewarping (Page Level))

So I downloaded the model ocrd resmgr with ocrd resmgr download ocrd-anybaseocr-dewarp '*' so far.
After runing ocrd-anybaseocr-dewarp -I OCR-D-DESKEW-PAGE -O OCR-D-DEWARP-PAGE

the following error message is returned:
[...] input_label = ``label_map.data.cuda()

File "/home/superman/ocrd_all/venv/local/sub-venv/headless-tf21/lib/python3.6/site-packages/torch/cuda/__init__.py", line 172, in _lazy_init torch._C._cuda_init()

RuntimeError: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver from http://www.nvidia.com/Download/index.aspx

[kba]

The parameters are from: [https://ocr-d.de/en/workflows] (look there at Step 6: Dewarping (Page Level))

Indeed, that is outdated, thanks for the heads-up, I'll fix it.

The dewarping does require a GPU and working CUDA setup. Do you have an NVIDIA graphic card in your machine? If so, you might want to install the CUDA toolkit. While it is a hard requirement for this processor, it's a massive speed boost for many other processors.

If you don't have a GPU, have a look at ocrd-cis-ocropy-dewarp which does support dewarping on line-level.

[Aysoltan]

Thanks for the hint. We will try to intigrate the high-performance GPU like NVIDIA later.

Now another thing: performing
ocrd-tesserocr-segment -I OCR-D-DESKEW-PAGE -O OCR-D-SEG -P find_tables false -P shrink_polygons true

resolves the following error message:

File "/home/superman/ocrd_all/venv/lib/python3.6/site-packages/ocrd_tesserocr/recognize.py", line 217, in process
    model = get_languages()[1][-1] # last installed model
IndexError: list index out of range

?

[kba]

Thanks for the hint. We will try to intigrate the high-performance GPU like NVIDIA later.

Now another thing: performing
ocrd-tesserocr-segment -I OCR-D-DESKEW-PAGE -O OCR-D-SEG -P find_tables false -P shrink_polygons true

resolves the following error message:

File "/home/superman/ocrd_all/venv/lib/python3.6/site-packages/ocrd_tesserocr/recognize.py", line 217, in process
    model = get_languages()[1][-1] # last installed model
IndexError: list index out of range

?

You need to have the osd and eng models installed as a minimum. Try

ocrd resmgr download ocrd-tesserocr-recognize '*'

which will install all the relevant models (don't worry, they are just a few MiB per model).

[kba]

BTW, you can call make install-models in the ocrd_all repo root folder to install all the models for all the processors in one go and not have to install them piece-by-piece later on.

[Aysoltan]

Many thanks for your hints. Everything went well. Is there also some commad for downloadind cor-asv-ann-models for post-correction?

[kba]

Is there also some commad for downloadind cor-asv-ann-models for post-correction?

No, I think they are only available via Git LFS which we haven't implemented an integration for. @bertsky?

[bertsky]

Many thanks for your hints. Everything went well. Is there also some commad for downloadind cor-asv-ann-models for post-correction?

@Aysoltan
No, we have not integrated these into the ocrd resmgr mechanism yet. You need to download them from the Github webpage. Or better yet (because our LFS quota there is very small), from our Gitlab webpage, which it is a mirror of.

For more information on the models, please contact me via Gitter (same nick as here) or open an issue on https://github.com/ASVLeipzig/cor-asv-ann.