Running RapidOCR with GPU Acceleration

[BennisonDevadoss]

I've been using RapidOCR for text extraction from images, and it's working well. However, I noticed that by default, it processes images using the CPU. I would like to know if it's possible to run RapidOCR with GPU acceleration for better performance.

Here's the relevant part of my code:

def read_text_from_images(image_paths):
    engine = RapidOCR()
    for img_path in image_paths:
        result, elapse = engine(img_path)
        print(f"Results for {img_path}: {result}")
        print(f"Time taken: {elapse} seconds")

I'm using the following version of RapidOCR:

rapidocr-onnxruntime = 1.3.24

I tried initializing RapidOCR with use_gpu=True, but it still runs on the CPU. Has anyone successfully run RapidOCR with GPU acceleration? If so, could you please guide me on how to enable it?

Thanks in advance for any help!

---

[SWHL]

If you want to use GPU accelerated inference, it is recommended to use the rapidocr_paddle library. For detailed tutorials, see: docs

[BennisonDevadoss]

Thank you, @SWHL! I noticed that in the default config.yml file, there's an option to set the cuda flag to true. Would enabling it work for GPU-accelerated inference in this case?

I've attached a sample image of the file for reference and also shared the link: config.yaml.

In this case, should I switch to the rapidocr_paddle library instead?

[SWHL]

The cuda flag is a legacy from the previous attempt at onnxruntime-gpu acceleration, which was later found to be not very good.

If you want to use GPU acceleration, I recommend using rapidocr_paddle, see link for details.