pdf_images.py: command not found

[pverkind]

When trying to running acdc with a pdf instead of images, I immediately get the following error:

(mkdir -p images/0049Gotha.OrA1521; cd images/0049Gotha.OrA1521; bash -c "pdf_images.py ../../pdf/0049Gotha.OrA1521.pdf") && touch images/0049Gotha.OrA1521/_SUCCESS
bash: pdf_images.py: command not found
make: *** [Makefile:64: images/0049Gotha.OrA1521/_SUCCESS] Error 127

The script creates a folder for the images extracted from the pdf but then fails to recognize that the pdf_images.py is in the bin folder of the acdc_train folder.

[pverkind]

hacky solution: I included the absolute path to the pdf_images.py file in the makefile

[pverkind]

I now have the same issue with the other files in the bin folder. I solved it too by creating a variable with the absolute path to these scripts in the makefile, but that should not be necessary.
This is on Ubuntu, Python 3.10, in a venv virtual environment.

[rohanchn]

Were you able to test the whole thing?

After the changes I made to the makefile, I was able to run it with the original command, however the $(ocr) variable in the final rule keeps on using the base model for subsequent training iterations intead of updating to the fine-tuned model from the previous iteration.

It should be doing the latter, right?

Edit: I guess it is doing the former as the latter might lead to overfitting.

[rohanchn]

Also, still not clear to me why gen2-%.out/alto-union/alto.lis uses lines from both print, gen1-print, and gen2-print?

Wouldn't there be duplicate lines in the final iteration because of this?