why yolov3-spp use maxpool size=5,9,13?

pjreddie / darknet

Convolutional Neural Networks

http://pjreddie.com/darknet/

Other

25.7k stars 21.33k forks source link

why yolov3-spp use maxpool size=5,9,13? #1529

Open libralx opened 5 years ago

libralx commented 5 years ago

What are the advantages？ why not use the spp-net default 11,22,4*4?

libralx commented 5 years ago

1×1， 2×2 ， 4×4

qwe3208620 commented 4 years ago

I think if you can compare mAP in both cases, you should be able to draw conclusions

zugofn commented 3 years ago

SPP is for get global and local feature. YOLOV3-SPP's SPP model is used for the feature map which size is 19x19. So for get the global feature,the kernel size(eg,13x13) should be close to the feature map size(19x19)