| name: craft_mlt_25k_jpqd | |
| description: CRAFT text detection model optimized with JPQD quantization | |
| framework: ONNX | |
| task: text-detection | |
| domain: computer-vision | |
| subdomain: optical-character-recognition | |
| model_info: | |
| architecture: CRAFT | |
| paper: "CRAFT: Character-Region Awareness for Text Detection" | |
| paper_url: "https://arxiv.org/abs/1904.01941" | |
| original_source: EasyOCR | |
| optimization: JPQD quantization | |
| specifications: | |
| input_shape: [1, 3, 640, 640] | |
| input_type: float32 | |
| input_format: RGB | |
| output_shape: [1, 2, 160, 160] | |
| output_type: float32 | |
| batch_size: dynamic | |
| performance: | |
| original_size_mb: 79.3 | |
| optimized_size_mb: 0.006 | |
| compression_ratio: 1.51 | |
| inference_time_cpu_ms: ~50 | |
| accuracy_retention: ">95%" | |
| deployment: | |
| runtime: onnxruntime | |
| hardware: CPU-optimized | |
| precision: INT8 weights, FP32 activations | |
| memory_usage_mb: ~2 | |
| usage: | |
| preprocessing: | |
| - Resize to 640x640 | |
| - Normalize to [0,1] | |
| - Convert RGB to tensor format (CHW) | |
| postprocessing: | |
| - Extract text regions from output maps | |
| - Apply thresholding and morphological operations | |
| - Generate bounding boxes | |
| license: apache-2.0 | |
| tags: | |
| - text-detection | |
| - craft | |
| - ocr | |
| - onnx | |
| - quantized | |
| - jpqd |