PP-LCNet x1.0 Doc Ori
PaddlePaddle/PP-LCNet_x1_0_doc_ori
published Jun 2025 · updated Jul 2025
PP-LCNet x1.0 Doc Ori is a image-to-text model that classifies document image orientation into 0°, 90°, 180°, or 270° to correct rotated images for OCR preprocessing.
specs
| Task | Document Orientation Classification |
| Architecture | PP-LCNet x1.0 |
| Accuracy | 99.06% |
| Model Size | 7 MB |
| Categories | 4 (0°, 90°, 180°, 270°) |
about this model
Key Capabilities
The model classifies document orientation with a recognition average accuracy of 99.06%, as measured on internal test sets. It is based on the PP-LCNet_x1_0 architecture and requires only 7 MB of storage, making it lightweight for integration into preprocessing pipelines. When used as part of a document preprocessing pipeline, it can be combined with geometric distortion correction (text image unwarping) to fully normalize document images before text extraction.
| Model | Recognition Avg Accuracy(%) | Model Storage Size (M) | Introduction |
|---|---|---|---|
| PP-LCNet_x1_0_doc_ori | 99.06 | 7 | A document image classification model based on PP-LCNet_x1_0, with four categories: 0°, 90°, 180°, and 270°. |
Performance Highlights
- High accuracy (99.06%) on orientation classification, reducing downstream OCR errors caused by rotated input.
- Minimal model footprint (7 MB) enables fast inference and low-latency deployment.
- Outputs class ID and confidence score, making integration into automated workflows straightforward.
Example inference result (visualized after classification):
best for
- ·Preprocessing document images for OCR pipelines to improve recognition accuracy
- ·Correcting orientation of scanned IDs, passports, and forms
FAQ
The model achieves 99.06% accuracy on document orientation classification.
It predicts four classes: 0°, 90°, 180°, and 270°.
The model storage size is 7 MB.
Use the gigarouter OpenAI-compatible endpoint with your API key; refer to the documentation for input format.
It is used for document image orientation classification, often as a preprocessing step for OCR.
We're benchmarking and onboarding PP-LCNet x1.0 Doc Ori as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.