I’ve been using the deepstream-app for object detection using YoloV3. When I use the regular darknet YoloV3 config file everything is fine, but recently I’ve been trying to make some customisations and I’m getting an error saying: “WARNING: Num classes mismatch. Configured:4, detected by network: 80”, followed by a segmentation fault.
The change I made to the yolov3.cfg file were the following:
- set layers = -1, 11 (instead of https://github.com/AlexeyAB/darknet/blob/6390a5a2ab61a0bdf6f1a9a6b4a739c16b36e0d7/cfg/yolov3.cfg#L720)
- set stride = 4 (instead of https://github.com/AlexeyAB/darknet/blob/6390a5a2ab61a0bdf6f1a9a6b4a739c16b36e0d7/cfg/yolov3.cfg#L717)
Would this kind of change require creation of a custom plugin(s)?
I’m not sure why the classes mismatch is occurring. All the [yolo] layers in the config have the correct number of classes. I’ve also edited the config_infer_primary_yoloV3.txt to have the same number of classes, and the labels file as well.