config yolov3-spp file for training for both small and large objects

[zhongjiejiang]

Hi AlexeyAB,

I download the yolov3-spp.cfg from below link, and I have some question to trouble you.

1, I found that the batch=1 and subdivisons=1, is it special mean? Can I set the batch=63 and subdivsion=8? 2, can I recalculate anchors for dataset? the smallest anchors of yolov3-spp is 10,13 and the smallest anchors of recalculate anchors is 22,39 3, how can it get he pre-train weight like as yolov3-tiny.conv.15 using command: darknet.exe partial?

Thanks for you reply Jason

for training for both small and large objects use modified models:

• Full-model: 5 yolo layers: https://raw.githubusercontent.com/AlexeyAB/darknet/master/cfg/yolov3_5l.cfg
• Tiny-model: 3 yolo layers: https://raw.githubusercontent.com/AlexeyAB/darknet/master/cfg/yolov3-tiny_3l.cfg
• Spatial-full-model: 3 yolo layers: https://raw.githubusercontent.com/AlexeyAB/darknet/master/cfg/yolov3-spp.cfg

[AlexeyAB]

@zhongjiejiang Hi

1. You can use any batch and subdivisions as you want. It will affect only on training.

2. Yes, you can recalculate anchors, do it as usual.

3. darknet.exe partial cfg/yolov3-spp.cfg yolov3-spp.weights yolov3-spp.conv.87 87