Nothing is detected after replacing onnx

[tingriley]

Current Behavior

looks like nothing has been deteced when using a different onnx file

How to Reproduce

1. download yolov4 pre-trained weights from https://github.com/kiyoshiiriemon/yolov4_darknet
2. convert weights to onnx using ./scripts/yolo2onnx.py --config yolov4.cfg --weights yolov4.weights
3. run detection python3 ./app.py -i ~/benchmark/MOT20-01-raw.mp4 --mot -o test.mp4 [code in yolo.py] ENGINE_PATH = Path(file).parent / 'yolov4.trt' MODEL_PATH = Path(file).parent / 'yolov4.onnx' NUM_CLASSES = 80 INPUT_SHAPE = (3, 608, 608) LAYER_FACTORS = [8, 16, 32] SCALES = [1.2, 1.1, 1.05] ANCHORS = [[12, 16, 19, 36, 40, 28],[36, 75, 76, 55, 72, 146],[142, 110, 192, 243, 459, 401]]

Describe what you want to do

1. What input videos you will provide, if any: https://motchallenge.net/vis/MOT20-01
2. What outputs you are expecting: detect at least one class of objects
3. Ask your questions here, if any: the program runs perfectly when using yolov4_crowdhuman.onnx, however, it fails to detect anything when using custom cfg and weights. Is there anything I did wrong? Any feedback would be appreciated.

Your Environment

• Desktop
  • Operating System and version: ubuntu18.04
  • NVIDIA Driver version: cuda_10.2
  • Used the docker image? NO

[tingriley]

I solved it by adding class_ids to mot_cfg.json