craft-mlt-25k-onnx

This repository contains an optimized ONNX version of the CRAFT (Character Region Awareness for Text) model.

Model Details

• Architecture: CRAFT (VGG16-based)
• Format: ONNX (Single File)
• Opset: 18
• Task: Scene Text Detection

Input Requirements

• Shape: [batch_size, 3, height, width]
• Constraints: height and width must be multiples of 32.
• Normalization: Scale pixels to [0, 1] (divide by 255).

Usage

import onnxruntime as ort
from huggingface_hub import hf_hub_download

# Download and load model
model_path = hf_hub_download(repo_id="KvaytG/craft-mlt-25k-onnx", filename="craft.onnx")
session = ort.InferenceSession(model_path, providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])

# Prepare input (example)
# img_tensor shape: (1, 3, 768, 768)
outputs = session.run(None, {'input': img_tensor})

Credits

Original model by Clova AI Research.

License

Licensed under the MIT license.