I was just wondering if it is possible to amend the yolo cfg file (which I assume is inside the container) in order to tune it for small objects (as alexeyab suggests in GitHub - AlexeyAB/darknet: YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet ) (ie change 3 lines defining layers and strides)?
For YOLOv4, see YOLOv4 — TAO Toolkit 3.21.11 documentation .
kmeans command (
tao yolo_v4 kmeans ) to determine the best anchor shapes for your dataset and put those anchor shapes in the spec file
Many thanks for the swift reply. Yes, I’ve made those changes after finding the optimal anchor sizes - but the specific changes I’m referring to lie within the yolo cfg file:
or training for small objects (smaller than 16x16 after the image is resized to 416x416) - set layers = 23 instead of https://github.com/AlexeyAB/darknet/blob/6f718c257815a984253346bba8fb7aa756c55090/cfg/yolov4.cfg#L895
set stride=4 instead of https://github.com/AlexeyAB/darknet/blob/6f718c257815a984253346bba8fb7aa756c55090/cfg/yolov4.cfg#L892
set stride=4 instead of https://github.com/AlexeyAB/darknet/blob/6f718c257815a984253346bba8fb7aa756c55090/cfg/yolov4.cfg#L989
Do you know whether this is possible to change in the tao implementation, and if so where the file is?
Thanks for the info. Above changes are not supported in the configuration.
I will sync internally for your request.