caffe python 接口api

目录

• caffe 某层featuremap ，和权重
• 只有deply，直接输出caffemodel
• 保存caffe权重到pkl文件
• 提取到的caffe权重pkl文件加载到pytorch中：
• caffemodel 修改与保存

caffe 某层featuremap ，和权重

    featuremap_1 = net.blobs['conv1_1'].data   #某一层feature map 输出
    filters = net.params['conv1_1'][0].data    #某一层权重

只有deply，直接输出caffemodel

net = caffe.Net(path_deply, phase=caffe.TEST)
net.save("./111.caffemodel")

保存caffe权重到pkl文件

import pickle as pkl
#feature map and shape
    print("=================feature map===================")
    for layer_name, blob in net.blobs.iteritems():
        print layer_name + '\t' + str(blob.data.shape)

    print("=================weights===================")
    for layer_name, blob in net.params.iteritems():
        len_ = len(blob)
        print layer_name + " has " + str(len_) + " params"
        for i in range(len_):
            print layer_name + ' idx=  '  + str(i) + '\t' + str(blob[i].data.shape)


    ############################################################################################3
    name_weights = {}
    # 保存每层的参数信息
    keys = open('keys.txt', 'w')
    keys.write('generated by Net-Caffe/convert_to_pkl.py\n\n')
    # 遍历每一网络层
    for param_name in net.params.keys():
        name_weights[param_name] = {}
        # 得到此层的参数
        layer_params = net.params[param_name]
        if len(layer_params) == 1:
            # 如果参数只有一个，则说明是反卷积层,
            # SfSNet整个模型里就只有反卷积层只有一组weight参数
            weight = layer_params[0].data
            name_weights[param_name]['weight'] = weight

            print('%s:\n\t%s (weight)' % (param_name, weight.shape))
            keys.write('%s:\n\t%s (weight)\n' % (param_name, weight.shape))
        elif len(layer_params) == 2:
            # 如果参数有两个，则说明是卷积层或者全连接层。
            # 卷积层或者全连接层都有两组参数：weight和bias
            # 权重参数
            weight = layer_params[0].data
            name_weights[param_name]['weight'] = weight
            # 偏置参数
            bias = layer_params[1].data
            name_weights[param_name]['bias'] = bias

            print('%s:\n\t%s (weight)' % (param_name, weight.shape))
            print('\t%s (bias)' % str(bias.shape))
            keys.write('%s:\n\t%s (weight)\n' % (param_name, weight.shape))
            keys.write('\t%s (bias)\n' % str(bias.shape))
        elif len(layer_params) == 5:
            # 如果有三个，则说明是BatchNorm层。
            # BN层共有三个参数，分别是：running_mean、running_var和一个缩放参数。
            running_mean = layer_params[0].data  # running_mean
            name_weights[param_name]['running_mean'] = running_mean / layer_params[2].data
            running_var = layer_params[1].data  # running_var
            name_weights[param_name]['running_var'] = running_var / layer_params[2].data

            aa = layer_params[2].data
            print("bn=",aa)

            name_weights[param_name]['weight'] = layer_params[3].data
            name_weights[param_name]['bias'] = layer_params[4].data

            print('%s:\n\t%s (running_var)' % (param_name, running_var.shape),)
            print('\t%s (running_mean)' % str(running_mean.shape))
            keys.write('%s:\n\t%s (running_var)\n' % (param_name, running_var.shape))
            keys.write('\t%s (running_mean)\n' % str(running_mean.shape))
            keys.write('\t%s (weight)\n' % str(layer_params[3].data.shape))
            keys.write('\t%s (bias)\n' % str(layer_params[4].data.shape))
        else:
            # 如果报错，大家要检查自己模型哈
            raise RuntimeError("还有参数个数超过3个的层，别漏了兄dei！！！\n")
    keys.close()
    # 保存name_weights
    with open('weights20220420.pkl', 'wb') as f:
        pkl.dump(name_weights, f, protocol=2)

提取到的caffe权重pkl文件加载到pytorch中：

import torch
import torchvision
import pickle as pkl
from torch import nn
import torch.nn.functional as F
import pickle as pkl
import cv2
import numpy as np


if __name__ == '__main__':
    net = net()
    # net.eval()

    index = 0
    print("*" * 50)
    for name, param in list(net.named_parameters()):
        print(str(index) + ':', name, param.size())
        index += 1
    print("*" * 50)

    for k, v in net.state_dict().items():
        print(k)
        print(v.shape)
        # print(k,v)
    print("@" * 50)





    from torch import from_numpy
    with open('/media/algo/data_1/project/net_my_pytorch/bk/0420all_tiqu_convet/weights20220420.pkl', 'rb') as wp:
        name_weights = pkl.load(wp,encoding='latin1')#pickle py2 save   and py3 read
        state_dict = {}
        # aaa = name_weights['Scale1']['weight']
        # aa = from_numpy(name_weights['Scale1']['weight']).reshape([1,3,1,1])
        # print(aa)
        # state_dict['Scale1.scale'] = from_numpy(name_weights['Scale1']['weight']).reshape([1,3,1,1])
        #conv1_1
        state_dict['conv1_1.weight'] = from_numpy(name_weights['conv1_1']['weight'])
        state_dict['conv1_1.bias'] = from_numpy(name_weights['conv1_1']['bias'])
        #conv1_1_bn
        state_dict['conv1_1_bn.running_var'] = from_numpy(name_weights['conv1_1/bn']['running_var'])
        state_dict['conv1_1_bn.running_mean'] = from_numpy(name_weights['conv1_1/bn']['running_mean'])
        state_dict['conv1_1_bn.weight'] = from_numpy(name_weights['conv1_1/bn']['weight'])
        state_dict['conv1_1_bn.bias'] = from_numpy(name_weights['conv1_1/bn']['bias'])

        # conv1_2
        state_dict['conv1_2.weight'] = from_numpy(name_weights['conv1_2']['weight'])
        state_dict['conv1_2.bias'] = from_numpy(name_weights['conv1_2']['bias'])
        # conv1_2_bn
        state_dict['conv1_2_bn.running_var'] = from_numpy(name_weights['conv1_2/bn']['running_var'])
        state_dict['conv1_2_bn.running_mean'] = from_numpy(name_weights['conv1_2/bn']['running_mean'])
        state_dict['conv1_2_bn.weight'] = from_numpy(name_weights['conv1_2/bn']['weight'])
        state_dict['conv1_2_bn.bias'] = from_numpy(name_weights['conv1_2/bn']['bias'])

        # conv2_1
        state_dict['conv2_1.weight'] = from_numpy(name_weights['conv2_1']['weight'])
        state_dict['conv2_1.bias'] = from_numpy(name_weights['conv2_1']['bias'])
        # conv2_1_bn
        state_dict['conv2_1_bn.running_var'] = from_numpy(name_weights['conv2_1/bn']['running_var'])
        state_dict['conv2_1_bn.running_mean'] = from_numpy(name_weights['conv2_1/bn']['running_mean'])
        state_dict['conv2_1_bn.weight'] = from_numpy(name_weights['conv2_1/bn']['weight'])
        state_dict['conv2_1_bn.bias'] = from_numpy(name_weights['conv2_1/bn']['bias'])

        # conv2_2
        state_dict['conv2_2.weight'] = from_numpy(name_weights['conv2_2']['weight'])
        state_dict['conv2_2.bias'] = from_numpy(name_weights['conv2_2']['bias'])
        # conv2_2_bn
        state_dict['conv2_2_bn.running_var'] = from_numpy(name_weights['conv2_2/bn']['running_var'])
        state_dict['conv2_2_bn.running_mean'] = from_numpy(name_weights['conv2_2/bn']['running_mean'])
        state_dict['conv2_2_bn.weight'] = from_numpy(name_weights['conv2_2/bn']['weight'])
        state_dict['conv2_2_bn.bias'] = from_numpy(name_weights['conv2_2/bn']['bias'])


        ############################
        # conv3_1
        state_dict['conv3_1.weight'] = from_numpy(name_weights['conv3_1']['weight'])
        state_dict['conv3_1.bias'] = from_numpy(name_weights['conv3_1']['bias'])
        # conv3_1_bn
        state_dict['conv3_1_bn.running_var'] = from_numpy(name_weights['conv3_1/bn']['running_var'])
        state_dict['conv3_1_bn.running_mean'] = from_numpy(name_weights['conv3_1/bn']['running_mean'])
        state_dict['conv3_1_bn.weight'] = from_numpy(name_weights['conv3_1/bn']['weight'])
        state_dict['conv3_1_bn.bias'] = from_numpy(name_weights['conv3_1/bn']['bias'])

        # conv3_2
        state_dict['conv3_2.weight'] = from_numpy(name_weights['conv3_2']['weight'])
        state_dict['conv3_2.bias'] = from_numpy(name_weights['conv3_2']['bias'])
        # conv3_2_bn
        state_dict['conv3_2_bn.running_var'] = from_numpy(name_weights['conv3_2/bn']['running_var'])
        state_dict['conv3_2_bn.running_mean'] = from_numpy(name_weights['conv3_2/bn']['running_mean'])
        state_dict['conv3_2_bn.weight'] = from_numpy(name_weights['conv3_2/bn']['weight'])
        state_dict['conv3_2_bn.bias'] = from_numpy(name_weights['conv3_2/bn']['bias'])

        ############################
        # conv4_1
        state_dict['conv4_1.weight'] = from_numpy(name_weights['conv4_1']['weight'])
        state_dict['conv4_1.bias'] = from_numpy(name_weights['conv4_1']['bias'])
        # conv3_1_bn
        state_dict['conv4_1_bn.running_var'] = from_numpy(name_weights['conv4_1/bn']['running_var'])
        state_dict['conv4_1_bn.running_mean'] = from_numpy(name_weights['conv4_1/bn']['running_mean'])
        state_dict['conv4_1_bn.weight'] = from_numpy(name_weights['conv4_1/bn']['weight'])
        state_dict['conv4_1_bn.bias'] = from_numpy(name_weights['conv4_1/bn']['bias'])

        # conv3_2
        state_dict['conv4_2.weight'] = from_numpy(name_weights['conv4_2']['weight'])
        state_dict['conv4_2.bias'] = from_numpy(name_weights['conv4_2']['bias'])
        # conv3_2_bn
        state_dict['conv4_2_bn.running_var'] = from_numpy(name_weights['conv4_2/bn']['running_var'])
        state_dict['conv4_2_bn.running_mean'] = from_numpy(name_weights['conv4_2/bn']['running_mean'])
        state_dict['conv4_2_bn.weight'] = from_numpy(name_weights['conv4_2/bn']['weight'])
        state_dict['conv4_2_bn.bias'] = from_numpy(name_weights['conv4_2/bn']['bias'])

        ############################
        # conv5_1
        state_dict['conv5_1.weight'] = from_numpy(name_weights['conv5_1']['weight'])
        state_dict['conv5_1.bias'] = from_numpy(name_weights['conv5_1']['bias'])
        # conv5_1_bn
        state_dict['conv5_1_bn.running_var'] = from_numpy(name_weights['conv5_1/bn']['running_var'])
        state_dict['conv5_1_bn.running_mean'] = from_numpy(name_weights['conv5_1/bn']['running_mean'])
        state_dict['conv5_1_bn.weight'] = from_numpy(name_weights['conv5_1/bn']['weight'])
        state_dict['conv5_1_bn.bias'] = from_numpy(name_weights['conv5_1/bn']['bias'])

        # conv5_2
        state_dict['conv5_2.weight'] = from_numpy(name_weights['conv5_2']['weight'])
        state_dict['conv5_2.bias'] = from_numpy(name_weights['conv5_2']['bias'])
        # conv5_2_bn
        state_dict['conv5_2_bn.running_var'] = from_numpy(name_weights['conv5_2/bn']['running_var'])
        state_dict['conv5_2_bn.running_mean'] = from_numpy(name_weights['conv5_2/bn']['running_mean'])
        state_dict['conv5_2_bn.weight'] = from_numpy(name_weights['conv5_2/bn']['weight'])
        state_dict['conv5_2_bn.bias'] = from_numpy(name_weights['conv5_2/bn']['bias'])

        ############################
        # conv6_1
        state_dict['conv6_1.weight'] = from_numpy(name_weights['conv6_1']['weight'])
        state_dict['conv6_1.bias'] = from_numpy(name_weights['conv6_1']['bias'])
        # conv6_1_bn
        state_dict['conv6_1_bn.running_var'] = from_numpy(name_weights['conv6_1/bn']['running_var'])
        state_dict['conv6_1_bn.running_mean'] = from_numpy(name_weights['conv6_1/bn']['running_mean'])
        state_dict['conv6_1_bn.weight'] = from_numpy(name_weights['conv6_1/bn']['weight'])
        state_dict['conv6_1_bn.bias'] = from_numpy(name_weights['conv6_1/bn']['bias'])

        # conv6_2
        state_dict['conv6_2.weight'] = from_numpy(name_weights['conv6_2']['weight'])
        state_dict['conv6_2.bias'] = from_numpy(name_weights['conv6_2']['bias'])
        # conv6_2_bn
        state_dict['conv6_2_bn.running_var'] = from_numpy(name_weights['conv6_2/bn']['running_var'])
        state_dict['conv6_2_bn.running_mean'] = from_numpy(name_weights['conv6_2/bn']['running_mean'])
        state_dict['conv6_2_bn.weight'] = from_numpy(name_weights['conv6_2/bn']['weight'])
        state_dict['conv6_2_bn.bias'] = from_numpy(name_weights['conv6_2/bn']['bias'])

        # upscore1
        state_dict['upscore1.weight'] = from_numpy(name_weights['upscore1']['weight'])

        ############################
        # conv7_1
        state_dict['conv7_1.weight'] = from_numpy(name_weights['conv7_1']['weight'])
        state_dict['conv7_1.bias'] = from_numpy(name_weights['conv7_1']['bias'])
        # conv7_1_bn
        state_dict['conv7_1_bn.running_var'] = from_numpy(name_weights['conv7_1/bn']['running_var'])
        state_dict['conv7_1_bn.running_mean'] = from_numpy(name_weights['conv7_1/bn']['running_mean'])
        state_dict['conv7_1_bn.weight'] = from_numpy(name_weights['conv7_1/bn']['weight'])
        state_dict['conv7_1_bn.bias'] = from_numpy(name_weights['conv7_1/bn']['bias'])

        # conv7_2
        state_dict['conv7_2.weight'] = from_numpy(name_weights['conv7_2']['weight'])
        state_dict['conv7_2.bias'] = from_numpy(name_weights['conv7_2']['bias'])
        # conv7_2_bn
        state_dict['conv7_2_bn.running_var'] = from_numpy(name_weights['conv7_2/bn']['running_var'])
        state_dict['conv7_2_bn.running_mean'] = from_numpy(name_weights['conv7_2/bn']['running_mean'])
        state_dict['conv7_2_bn.weight'] = from_numpy(name_weights['conv7_2/bn']['weight'])
        state_dict['conv7_2_bn.bias'] = from_numpy(name_weights['conv7_2/bn']['bias'])

        # upscore2
        state_dict['upscore2.weight'] = from_numpy(name_weights['upscore2']['weight'])

        ############################
        # conv8_1
        state_dict['conv8_1.weight'] = from_numpy(name_weights['conv8_1']['weight'])
        state_dict['conv8_1.bias'] = from_numpy(name_weights['conv8_1']['bias'])
        # conv8_1_bn
        state_dict['conv8_1_bn.running_var'] = from_numpy(name_weights['conv8_1/bn']['running_var'])
        state_dict['conv8_1_bn.running_mean'] = from_numpy(name_weights['conv8_1/bn']['running_mean'])
        state_dict['conv8_1_bn.weight'] = from_numpy(name_weights['conv8_1/bn']['weight'])
        state_dict['conv8_1_bn.bias'] = from_numpy(name_weights['conv8_1/bn']['bias'])

        # conv8_2
        state_dict['conv8_2.weight'] = from_numpy(name_weights['conv8_2']['weight'])
        state_dict['conv8_2.bias'] = from_numpy(name_weights['conv8_2']['bias'])
        # conv8_2_bn
        state_dict['conv8_2_bn.running_var'] = from_numpy(name_weights['conv8_2/bn']['running_var'])
        state_dict['conv8_2_bn.running_mean'] = from_numpy(name_weights['conv8_2/bn']['running_mean'])
        state_dict['conv8_2_bn.weight'] = from_numpy(name_weights['conv8_2/bn']['weight'])
        state_dict['conv8_2_bn.bias'] = from_numpy(name_weights['conv8_2/bn']['bias'])

        # upscore3
        state_dict['upscore3.weight'] = from_numpy(name_weights['upscore3']['weight'])

        ############################
        # conv9_1
        state_dict['conv9_1.weight'] = from_numpy(name_weights['conv9_1']['weight'])
        state_dict['conv9_1.bias'] = from_numpy(name_weights['conv9_1']['bias'])
        # conv9_1_bn
        state_dict['conv9_1_bn.running_var'] = from_numpy(name_weights['conv9_1/bn']['running_var'])
        state_dict['conv9_1_bn.running_mean'] = from_numpy(name_weights['conv9_1/bn']['running_mean'])
        state_dict['conv9_1_bn.weight'] = from_numpy(name_weights['conv9_1/bn']['weight'])
        state_dict['conv9_1_bn.bias'] = from_numpy(name_weights['conv9_1/bn']['bias'])

        # conv9_2
        state_dict['conv9_2.weight'] = from_numpy(name_weights['conv9_2']['weight'])
        state_dict['conv9_2.bias'] = from_numpy(name_weights['conv9_2']['bias'])
        # conv9_2_bn
        state_dict['conv9_2_bn.running_var'] = from_numpy(name_weights['conv9_2/bn']['running_var'])
        state_dict['conv9_2_bn.running_mean'] = from_numpy(name_weights['conv9_2/bn']['running_mean'])
        state_dict['conv9_2_bn.weight'] = from_numpy(name_weights['conv9_2/bn']['weight'])
        state_dict['conv9_2_bn.bias'] = from_numpy(name_weights['conv9_2/bn']['bias'])

        #######################3
        #net_my_cc_loc
        state_dict['net_my_cc_loc.weight'] = from_numpy(name_weights['net_my_cc_loc']['weight'])
        state_dict['net_my_cc_loc.bias'] = from_numpy(name_weights['net_my_cc_loc']['bias'])
        # net_my_cc_conf
        state_dict['net_my_cc_conf.weight'] = from_numpy(name_weights['net_my_cc_conf']['weight'])
        state_dict['net_my_cc_conf.bias'] = from_numpy(name_weights['net_my_cc_conf']['bias'])
        # net_my_cc1_loc
        state_dict['net_my_cc1_loc.weight'] = from_numpy(name_weights['net_my_cc1_loc']['weight'])
        state_dict['net_my_cc1_loc.bias'] = from_numpy(name_weights['net_my_cc1_loc']['bias'])
        # net_my_cc1_conf
        state_dict['net_my_cc1_conf.weight'] = from_numpy(name_weights['net_my_cc1_conf']['weight'])
        state_dict['net_my_cc1_conf.bias'] = from_numpy(name_weights['net_my_cc1_conf']['bias'])

        #######################3
        # net_my_rd_loc
        state_dict['net_my_rd_loc.weight'] = from_numpy(name_weights['net_my_rd_loc']['weight'])
        state_dict['net_my_rd_loc.bias'] = from_numpy(name_weights['net_my_rd_loc']['bias'])
        # net_my_rd_conf
        state_dict['net_my_rd_conf.weight'] = from_numpy(name_weights['net_my_rd_conf']['weight'])
        state_dict['net_my_rd_conf.bias'] = from_numpy(name_weights['net_my_rd_conf']['bias'])
        # net_my_rd1_loc
        state_dict['net_my_rd1_loc.weight'] = from_numpy(name_weights['net_my_rd1_loc']['weight'])
        state_dict['net_my_rd1_loc.bias'] = from_numpy(name_weights['net_my_rd1_loc']['bias'])
        # net_my_rd1_conf
        state_dict['net_my_rd1_conf.weight'] = from_numpy(name_weights['net_my_rd1_conf']['weight'])
        state_dict['net_my_rd1_conf.bias'] = from_numpy(name_weights['net_my_rd1_conf']['bias'])

        #######################3
        # net_my_pd_loc
        state_dict['net_my_pd_loc.weight'] = from_numpy(name_weights['net_my_pd_loc']['weight'])
        state_dict['net_my_pd_loc.bias'] = from_numpy(name_weights['net_my_pd_loc']['bias'])
        # net_my_pd_conf
        state_dict['net_my_pd_conf.weight'] = from_numpy(name_weights['net_my_pd_conf']['weight'])
        state_dict['net_my_pd_conf.bias'] = from_numpy(name_weights['net_my_pd_conf']['bias'])
        # net_my_pd1_loc
        state_dict['net_my_pd1_loc.weight'] = from_numpy(name_weights['net_my_pd1_loc']['weight'])
        state_dict['net_my_pd1_loc.bias'] = from_numpy(name_weights['net_my_pd1_loc']['bias'])
        # net_my_pd1_conf
        state_dict['net_my_pd1_conf.weight'] = from_numpy(name_weights['net_my_pd1_conf']['weight'])
        state_dict['net_my_pd1_conf.bias'] = from_numpy(name_weights['net_my_pd1_conf']['bias'])

        #######################3
        # net_my_vd_loc
        state_dict['net_my_vd_loc.weight'] = from_numpy(name_weights['net_my_vd_loc']['weight'])
        state_dict['net_my_vd_loc.bias'] = from_numpy(name_weights['net_my_vd_loc']['bias'])
        # net_my_vd_conf
        state_dict['net_my_vd_conf.weight'] = from_numpy(name_weights['net_my_vd_conf']['weight'])
        state_dict['net_my_vd_conf.bias'] = from_numpy(name_weights['net_my_vd_conf']['bias'])
        # net_my_vd1_loc
        state_dict['net_my_vd1_loc.weight'] = from_numpy(name_weights['net_my_vd1_loc']['weight'])
        state_dict['net_my_vd1_loc.bias'] = from_numpy(name_weights['net_my_vd1_loc']['bias'])
        # net_my_vd1_conf
        state_dict['net_my_vd1_conf.weight'] = from_numpy(name_weights['net_my_vd1_conf']['weight'])
        state_dict['net_my_vd1_conf.bias'] = from_numpy(name_weights['net_my_vd1_conf']['bias'])



    net.load_state_dict(state_dict)

    net.cuda()
    net.eval()

    torch.save(net.state_dict(), './net_my_model_20220420.pth')

caffemodel 修改与保存

path_deply_goe = "/media/algo/data_1/everyday//20220512weight/g_deply1.4.4.prototxt"
path_model_goe = "/media/algo/data_1/everyday/20220512weight/g.caffemodel"

path_deply_lld = "/media/algo/data_1/everyday//20220512weight/l_deply.prototxt"
path_model_lld = "/media/algo/data_1/everyday//20220512weight/l-3000.caffemodel"

path_backbone_deply = "/media/algo/data_1/everyday/20220512weight/backbone_deply.prototxt"

caffe.set_mode_cpu()
caffe.set_device(0)
##39
list_backbone_layer = ["conv1_1", "conv1_1/bn", "conv1_2", "conv1_2/bn", "conv2_1", "conv2_1/bn", "conv2_2",\
                       "conv2_2/bn", "conv3_1", "conv3_1/bn", "conv3_2", "conv3_2/bn", "conv4_1", "conv4_1/bn",\
                       "conv4_2", "conv4_2/bn", "conv5_1", "conv5_1/bn", "conv5_2", "conv5_2/bn", "conv6_1",\
                       "conv6_1/bn", "conv6_2", "conv6_2/bn", "upscore1", "conv7_1", "conv7_1/bn", "conv7_2",\
                       "conv7_2/bn", "upscore2", "conv8_1", "conv8_1/bn", "conv8_2", "conv8_2/bn", "upscore3",\
                       "conv9_1", "conv9_1/bn", "conv9_2", "conv9_2/bn"]

net_goe = caffe.Net(path_deply_goe, path_model_goe, caffe.TEST)
net_lld = caffe.Net(path_deply_lld, path_model_lld, caffe.TEST)
net_merge = caffe.Net(path_backbone_deply, phase=caffe.TEST)
for layer_name, para in net_goe.params.items():
    print("layer_name=", layer_name, "      :::len=", len(para))
    if layer_name in list_backbone_layer:
        for i in range(len(para)):
            weight_goe = net_goe.params[layer_name][i].data[...]  ## weights = para[i].data[...]
            weight_lld = net_lld.params[layer_name][i].data[...]
            weight_merge = weight_goe * 0.4 + weight_lld * 0.6
            net_merge.params[layer_name][i].data[...] = weight_merge

net_merge.save("./merge.caffemodel")