search

devmaxxing / videocr-PaddleOCR

Extract hardcoded subtitles from videos using machine learning
MIT License
142 stars 22 forks source link

Guide to installation on Docker? Is there a stable/tested build for a specific Paddle image? #11

Closed BrianZbr closed 1 year ago

BrianZbr commented 1 year ago

I'm having a lot of trouble getting this working. I thought I'd have better luck in Docker, but at this point I've tried both paddlepaddle/paddle:2.4.1 and paddlepaddle/paddle:2.3.1 and in both cases after I install videocr-PaddleOCR, the paddle package can no longer be imported (Illegal instruction (core dumped)). It seems like videocr-PaddleOCR and paddle are fighting over the versions required for various packages.

Is there a there a stable version of videocr-PaddleOCR which is known to work with any specific paddle Docker image?

devmaxxing commented 1 year ago

Does the Google Colab work for you? I have not tried installing on Docker yet.

The latest version of videocr-PaddleOCR should work with paddlepaddle 2.3.x The Google Colab notebook uses pip install paddlepaddle==2.3.1 -i https://mirror.baidu.com/pypi/simple for CPU and pip install paddlepaddle-gpu==2.3.1.post112 -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html for GPU.

BrianZbr commented 1 year ago

Thanks! I'd overlooked the notebook link until you mentioned it here. It might be worth highlighting that more and briefly explaining what it is in the README.

The notebook does work for me, but I'd have to figure out if I can stop it from timing out long enough to process a significant chunk of video. At least now that I know it works in a specific environment I can access, I might work on getting it set up in Docker.

I'll leave this issue open if you want to assign it to me? I'll make a PR if I have suggested updated to the README.

EDIT: Tested with a shorter video and the results were pretty disappointing! I may have had unrealistic hopes for Paddle but it seems like using Tesseract is close enough. I will close this out.

devmaxxing commented 1 year ago

what language are you working with? PaddleOCR has worked pretty well for me for English and Chinese. If you are using the colab notebook defaults, you'll probably want to adjust the crop window