Caffe学习-计算模型参数个数和FLOPs

个人微信公众号：AI研习图书馆，欢迎关注~

深度学习知识及资源分享，学习交流，共同进步~

1. 代码

import sys
sys.path.insert(0,"python")
import caffe
 
model="models/bvlc_alexnet/deploy.prototxt"
 
def main():
    net=caffe.Net(model,caffe.TEST)
    params=0
    flops=0
    blobs=net.blobs
    print("name param flops")
    for item in net.params.items():
        name,layer=item
        c1=layer[0].count
        c2=layer[1].count
        b=blobs[name]
        param=c1+c2
        flop=param*b.width*b.height
        print(name+" "+str(param)+" "+str(flop))
        params+=param
        flops+=flop
    print("total params",params)
    print("FLOPs:",flops)
if __name__ == '__main__':
    main()

2. 结果展示

name param flops
conv1 34944 105705600
conv2 307456 224135424
conv3 885120 149585280
conv4 663936 112205184
conv5 442624 74803456
fc6 37752832 37752832
fc7 16781312 16781312
fc8 4097000 4097000
('total params', 60965224)
('FLOPs:', 725066088)

您的支持，是我不断创作的最大动力~

欢迎点赞，关注，留言交流~

深度学习，乐此不疲~