tanluren / yolov3-channel-and-layer-pruning

yolov3 yolov4 channel and layer pruning, Knowledge Distillation 层剪枝,通道剪枝,知识蒸馏
Apache License 2.0
1.5k stars 446 forks source link

为什么prune.py跑出的map和之前train顺便跑出来的不一样? #41

Open GuoxingYan opened 4 years ago

GuoxingYan commented 4 years ago

train里的精度 image prune显示的精度,p下降了好多。

image +------------+----------+----------+ | Metric | Before | After | +------------+----------+----------+ | mAP | 0.882270 | 0.881940 | | Parameters | 61523734 | 7517869 | | Inference | 0.0431 | 0.0219 | +------------+----------+----------+

zbyuan commented 4 years ago

进行模型剪枝 最后一个epoch的conf是0.001,p肯定低 还有就是你的剪枝这里损失不大,map掉的不多,p掉了主要是因为conf不同

GuoxingYan commented 4 years ago

进行模型剪枝 最后一个epoch的conf是0.001,p肯定低 还有就是你的剪枝这里损失不大,map掉的不多,p掉了主要是因为conf不同

谢谢,我仔细看了下代码,确实是这样,一般情况是0.1,last epoch和剪枝的是0.001.那一般做精度评定的时候,以哪一个为准呢?

zhangnan-hust commented 3 years ago

请问为什么最后一个epoch要调整conf的阈值?

Luoxsh6 commented 3 years ago

请问为什么最后一个epoch要调整conf的阈值?

看代码解释是为了speedup