Number of anchors reducing mAP

forresti / SqueezeNet

SqueezeNet: AlexNet-level accuracy with 50x fewer parameters

BSD 2-Clause "Simplified" License

2.17k stars 723 forks source link

Number of anchors reducing mAP #40

Closed soyebn closed 6 years ago

soyebn commented 7 years ago

Any intuition why does increasing number of anchors from 9 to 16, decreases mAP by 10%? I was expecting if it does not increase mAP at least it should have kept same. Is it due to replacing FC with conv layer in your SqueezeDet vs. YOLO?

sunnyxiaohu commented 7 years ago

How do you set the layer of learning rate and initialization mode ? weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0 }

param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 }

?? How about the freeze_layers ?

forresti commented 6 years ago

This repo is SqueezeNet Are you looking for SqueezeDet? https://github.com/BichenWuUCB/squeezeDet