| --- |
| license: mit |
| pipeline_tag: image-to-text |
| tags: |
| - ocr |
| - ctc |
| - pytorch |
| - text-recognition |
| - lightweight |
| language: |
| - en |
| --- |
| |
| # QOCR-Tiny v1 |
|
|
| A compact CNN + BiGRU + CTC text recognition model. Built from scratch, no pretrained weights, released under MIT License. |
|
|
| ## Usage |
|
|
| ### Command line |
|
|
| ```bash |
| python inference.py path/to/image.png |
| ``` |
|
|
| Prints the recognized text to stdout. |
|
|
| ### Python |
|
|
| ```python |
| from inference import ocr_file |
| |
| text = ocr_file("path/to/image.png") |
| print(text) |
| ``` |
|
|
| Or, working directly with a PIL image: |
|
|
| ```python |
| from PIL import Image |
| from inference import ocr |
| |
| image = Image.open("path/to/image.png") |
| text = ocr(image) |
| print(text) |
| ``` |
|
|
| ## Limitations |
|
|
| - No text detection — you supply the region to read; the model does not locate text within a larger image. |
| - Greedy CTC decoding only; no beam search or language model rescoring. |
|
|
| ## License |
|
|
| MIT License. |
