use pytorch to reimplement paddle detection
This project aims to:
- Learn object detection and PPDetection code
- Enable PPDetection model to be used on pytorch
- Provide a reference for paddle to pytorch
- Coordinate transformation of target detection in yolov3 inference part
- Rotating frame detection model PP-YOLOE-R
- Small target detection model PP-YOLOE-SOD
- PP-Human v2
- PP-Vehicle
- Object detection model training part
- Multiple target tracking
- ...
PPDetectionPytorch is a pytorch implementation based on PaddleDetection and PaddleYOLO.
RECENT UPDATED
- 2023-05-31 fix issue #1 yolov6 infer update
- 2023-05-26 Migration model inference part: rt-detr(RT-DETR-R50-m,RT-DETR-R34,RT-DETR-R18)
- 2023-04-25 Migration model inference part: rt-detr(DETRs Beat YOLOs on Real-time Object Detection)
- 2023-02-13 Migration model inference part: retinanet
- 2023-02-11 Migration model inference part: blazeface
- 2023-02-03 Migration model inference part: ssd
- 2023-01-20 Migration model inference part: yolov8
- 2023-01-06 Coordinate transformation of target detection in yolov3 inference part
- 2022-12-21 Migration model inference part: picodet, ppyoloe, yolox, yolov5, yolov6, yolov7, rtmdet
- 2022-12-19 Base code migration
WORK_DIR Constants.WORK_DIR is changed to the absolute path of your own code
Automatically download the corresponding model weights in configs to the ~/.cache/paddle/weights directory during the first inference. This directory is the specified directory for downloading the PPDetection inference model. Model weights will be automatically converted to pytoch weights and saved to the current directory.
Model weight automatic conversion code
Currently only part of the model weights are converted: The run_picodet_coco_batch function in run_detection_infer.py uses different model_classes for reasoning tests
Output inference results and corresponding network paddle and pytorch model weight names Inference output directory