Handwritten digits detection using a YOLOv8 detection model and ONNX pre/post processing. An example of how model works in real world scenario can be viewed at https://thawro.github.io/web-object-detector/.
The dataset consists of images created with the use of a HWD+ dataset (more here).
Each pipeline step is done with ONNX models. The complete pipeline during inference is the following:
- Image preprocessing - resize and pad to match model input size (preprocessing)
- Object detection - Detect objects with YOLOv8 model (yolo)
- Non Maximum Supression - Apply NMS to YOLO output (nms)
- Postprocessing - Apply postprocessing to filtered boxes (postprocessing)
- PyTorch - neural networks architectures and datasets classes
- ONNX - All processing steps used in pipeline
- ONNX Runtime - Pipeline inference
- OpenCV - Image processing for the server-side model inference (optional)
- React - Web application used to test object detection models in real world examples
- Go to https://thawro.github.io/web-object-detector/
- Follow the instructions on the page