Created by: Fauny

模型：ppyolo 训练结果：Best test box ap: 0.7949662589940378 'loss': '26.862640' , 导出模型大小：188.1M。经测试模型比较满意 敏感度分析：获得敏感度分析文件 确定参数：使用一下代码获得

import math
from paddleslim.prune import load_sensitivities
from paddleslim.prune import get_ratios_by_loss 

sensitivities_file = "sstv.data"
sensitivities = load_sensitivities(sensitivities_file)
# print(sensitivities) 
ratios = get_ratios_by_loss(sensitivities, 0.1) 
print(ratios)

获得参数：

{
'yolo_block.0.0.0.conv.weights': 0.5024404642823515, 
'yolo_block.0.0.1.conv.weights': 0.5806873814522566, 
// 'yolo_block.0.1.0.conv.weights': 0.4130713279785401, 
'yolo_block.0.1.1.conv.weights': 0.8445627084858831, 
'yolo_block.0.2.conv.weights': 0.7140574970525724, 
'yolo_block.0.tip.conv.weights': 0.8026657765827588,
'yolo_block.1.0.0.conv.weights': 0.9,
'yolo_block.1.0.1.conv.weights': 0.8590961280425866, 
'yolo_block.1.1.0.conv.weights': 0.7435697192868094, 
'yolo_block.1.1.1.conv.weights': 0.8933719407087457, 
'yolo_block.1.2.conv.weights': 0.796605418059164, 
'yolo_block.1.tip.conv.weights': 0.9, 
'yolo_block.2.0.0.conv.weights': 0.8641740135486591, 
'yolo_block.2.0.1.conv.weights': 0.7584828344125396,
'yolo_block.2.1.0.conv.weights': 0.7780280316817834,
'yolo_block.2.1.1.conv.weights': 0.8247591955828633, 
'yolo_block.2.2.conv.weights': 0.7763518811427775,
'yolo_block.2.tip.conv.weights': 0.8543156017782207
}

参考 #1346 (closed)， 去掉 yolo_block.0.1.0.conv.weights 最终参数：

--pruned_params "yolo_block.0.0.0.conv.weights,yolo_block.0.0.1.conv.weights,yolo_block.0.1.1.conv.weights,yolo_block.0.2.conv.weights,yolo_block.0.tip.conv.weights,yolo_block.1.0.0.conv.weights,yolo_block.1.0.1.conv.weights,yolo_block.1.1.0.conv.weights,yolo_block.1.1.1.conv.weights,yolo_block.1.2.conv.weights,yolo_block.1.tip.conv.weights,yolo_block.2.0.0.conv.weights,yolo_block.2.0.1.conv.weights,yolo_block.2.1.0.conv.weights,yolo_block.2.1.1.conv.weights,yolo_block.2.2.conv.weights,yolo_block.2.tip.conv.weights" \
--pruned_ratios="0.5,0.5,0.8,0.7,0.8,0.9,0.8,0.7,0.8,0.7,0.9,0.8,0.7,0.7,0.8,0.7,0.8" \

问题来了，训练中途，在box ap：0.7086416361742713 时测试，发现结果很不正常，与剪枝前结果大相径庭。 请教原因：

1. get_ratios_by_loss 使用 0.1 参数是不是太大？
2. 使用参数中，除了yolo_block.0.1.0.conv.weights ，还有别的不能剪的吗？