OCR To TXT (Ollama + GLM OCR)

Simple verbose shell pipeline to OCR images sequentially by image time.

• Finds images in input_images (heic, webp, png, jpg, jpeg)
• Sorts by time priority: DateTimeOriginal -> CreateDate -> FileModifyDate
• Converts each image to jpg at 1280x720 (720p)
• Sends each image to local Ollama model (default: glm-ocr:latest)
• Saves per-image OCR text and combined OCR text

• ocr_to_txt.sh: main OCR script
• create_synthetic_test_data.sh: creates 2 synthetic test images

• ollama running locally at http://127.0.0.1:11434
• commands: curl, jq, exiftool, base64, convert (ImageMagick) or magick

./create_synthetic_test_data.sh ./input_images
./ocr_to_txt.sh ./input_images ./work_jpg_720 ./ocr_output glm-ocr:latest

• Per image text: ./ocr_output/txt/*.txt
• Combined text: ./ocr_output/all_ocr.txt

• OLLAMA_URL (default http://127.0.0.1:11434)
• OCR_TIMEOUT (default 45; use 0 for no timeout)
• OCR_NUM_PREDICT (default 128)
• OCR_NUM_CTX (default 2048)
• CONVERT_TIMEOUT (default 120)
• OCR_PROMPT (custom OCR prompt)