打开微信,使用扫一扫进入页面后,点击右上角菜单,
点击“发送给朋友”或“分享到朋友圈”完成分享
不使用均值方差,同一个模型,其他参数不变,输入不变,use_firstconv开启与不开启,输出结果不同。
网络结构:
input: "input" input_dim: 1 input_dim: 3 input_dim: 384 input_dim: 384 { name: "conv1" type: "Convolution" bottom: "input" top: "conv1" convolution_param { num_output: 64 bias_term: false stride: 2 pad_h: 3 pad_w: 3 kernel_h: 7 kernel_w: 7 } }
convert_quantized.ini:
[model] original_models_path = ./test_conv.prototxt save_model_path = ./test_conv_int8.prototxt [data] images_list_path = ../ref_images/ref_images.txt used_images_num = 10 [weights] original_weights_path = ./test_conv.caffemodel [preprocess] ;mean value order should be the same as filter channel order ; mean = 0, 0, 0 ;support single std and channel std, channel order should be the same as filter channel order ;std=1/255 ; std = 0.00392157 scale = 384,384 crop = 384,384 [config] quantize_op_list = Conv, FC, LRN use_firstconv = 1 ;customer configuration [custom] use_custom_preprocess = 0
文档中对use_firstconv是这么说的:
我的理解是,如果不做均值方差,use_firstconv是否开启应该不影响结果的吧?
热门帖子
精华帖子