easyocr-redactor

Redacts keywords from images

HOW TO USE
1.
Download checker_v3_gpu_easyocr.py

Download and install python (3.11 or higher), install torch and torchvision, example for 5060ti:

python -m pip install torch torchvision --index-url https://download.pytorch.org/whl/cu132

and lastly easyocr:

python -m pip install --upgrade easyocr opencv-python pillow numpy

• File folder structure should look like this:
• Videos
  • frames
    • pic_0001.png
    • pic_0002.png
    • ...
  • checker_v3_gpu_easyocr.py

To run code:

python checker_v3_gpu_easyocr.py

Output should look like this:

20755/21625 | frame_020755.png | Total: 2.33s | Read: 0.19s | OCR: 1.02s | Redact: 0.01s | Save: 1.09s | Redacted: 3 | FPS: 0.34 | Elapsed: 17h 7m 44s | ETA: 43m 4s

Example pictures:
Before

After

Processing speed of 4K image with 2x amd epyc 7742, 256GB of ram and 5060ti (16GB) is between 2-3 seconds per frame.
Processing speed of 4K image with 5 1600x, 16GB of ram and 1660ti is between 1.5-2 seconds per frame.

for multi pc setup use checker_v7_multi_gpu_easyocr_remote_worker.py.
And with this one you need to select the devices to use and set the global total workers and worker ids for every device you want to use.

Example with 2 gpus:
PS D:\AI Videos\xqc\test> python checker_v7_multi_gpu_easyocr_remote_worker.py

Available Devices: [cpu] Central Processing Unit
[0] NVIDIA GeForce RTX 5060 Ti
[1] NVIDIA GeForce RTX 5060 Ti

Select LOCAL device(s) to use (example: cpu or 0 or cpu,0 or all): 0,1

Enter TOTAL number of global workers across all machines (e.g., 3): 2
Enter the 2 global worker ID(s) for this machine (1 to 2, example: 1 or 1,2): 1,2

=== Startup ===
Frames directory: frames_upscaled
Total frames globally: 60064
Total global workers: 2
This machine handles worker IDs: [1, 2]
Resume mode: skipped 0 already-processed frames assigned to this machine.
Frames to process NOW on this machine: 60064
Local Device [0] (Global Worker 1): 30032 frame(s)
Local Device [1] (Global Worker 2): 30032 frame(s)
Processing mode: in-place overwrite
Press Ctrl+C to stop safely.

Other thing

If there are any coders who want a challenge, try to get the old paddle ocr code working.