Modification to cfg to find many small objects

[stephanecharette]

There are instructions in the readme about modifications to make for small objects:

for 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 and set stride=4 instead of https://github.com/AlexeyAB/darknet/blob/6f718c257815a984253346bba8fb7aa756c55090/cfg/yolov4.cfg#L989

But these instructions are for the full-sized YOLOv4 cfg file. Are there equivalent recommendations when looking for small objects with yolov4-tiny.cfg?

In my case, the images are 800x800, it needs to run on a Jetson Nano IoT, and the objects I need to detect are quite small. Here is an example crop of a 400x400 section that I have begun marking up for training:

[lzcchl]

hello, I see that your Jetson Nano can process yolov4-tiny-3l in realtime, that faster than me, what tricks you use? (1)nvdec? (2)tensorRT? (3)thread？or what for (4)(5)...?

[bulatnv]

Hello @stephanecharette, I see that you are trying to detect small objects. Have you tried this network #4213?