Full pipeline from model training to deployment on Nvidia AGX Xavier. The following statements take into account that the model was trained on BDD100k dataset.
1 - Train model through: bag-of-models repo. Then, Convert the final model to .weights file;
2 - yolov4-bdd-512.weights file in "yolo" directory and run from the root directory:
python3 yolo_to_onnx.py --model yolov4-bdd-512
3 - From the root directory run:
python3 onnx_to_tensorrt.py --model yolov4-bdd-512
4 - At this point we have on "yolo" directory: .cfg, .weights, .onnx and .trt files. Finally, to run the application from the "scripts" directory with:
python3 run_yolo.py --model yolov4-bdd-512 --category_num 10 --video_dev 0