search

yun-liu / RCF

Richer Convolutional Features for Edge Detection
Other
753 stars 259 forks source link

Is AutoCrop layer only used in training or also in inference ? can i remove Autocrop layer in prototxt file while doing Inference? #24

Closed athiselvam closed 6 years ago

athiselvam commented 6 years ago

Is AutoCrop layer only used in training or also in inference ? can i remove Autocrop layer in prototxt file while doing Inference?

yun-liu commented 6 years ago

@athiselvam AutoCrop layer is used in both the training and the inference. You can not remove the Autocrop layer directly, because it is used to align the upsampling maps obtained by Deconvolution layer. But you can replace it with the standard Crop layer in Caffe by setting the offset to 1, 2, 4, and 0 for conv2, conv3, conv4, and conv5, respectively.

athiselvam commented 6 years ago

Thank You so much for your reply. Can you please send me the above change Prototxt file. Since I'm new to caffe, help me do so.

yun-liu commented 6 years ago

@athiselvam 

name: "RCF"

input: "data"
input_shape {
  dim: 1
  dim: 3
  dim: 224
  dim: 224
}

layer { name: "conv1_1" type: "Convolution" bottom: "data" top: "conv1_1"
  convolution_param { num_output: 64 pad: 1 kernel_size: 3 } }
layer { name: "relu1_1" type: "ReLU" bottom: "conv1_1" top: "conv1_1" }
layer { name: "conv1_2" type: "Convolution" bottom: "conv1_1" top: "conv1_2"
  convolution_param { num_output: 64 pad: 1 kernel_size: 3 } }
layer { name: "relu1_2" type: "ReLU" bottom: "conv1_2" top: "conv1_2" }
layer { name: "pool1" type: "Pooling" bottom: "conv1_2" top: "pool1"
  pooling_param { pool: MAX kernel_size: 2 stride: 2 } }

layer { name: "conv2_1" type: "Convolution" bottom: "pool1" top: "conv2_1"
  convolution_param { num_output: 128 pad: 1 kernel_size: 3 } }
layer { name: "relu2_1" type: "ReLU" bottom: "conv2_1" top: "conv2_1" }
layer { name: "conv2_2" type: "Convolution" bottom: "conv2_1" top: "conv2_2"
  convolution_param { num_output: 128 pad: 1 kernel_size: 3 } }
layer { name: "relu2_2" type: "ReLU" bottom: "conv2_2" top: "conv2_2" }
layer { name: "pool2" type: "Pooling" bottom: "conv2_2" top: "pool2"
  pooling_param { pool: MAX kernel_size: 2 stride: 2 } }

layer { name: "conv3_1" type: "Convolution" bottom: "pool2" top: "conv3_1"
  convolution_param { num_output: 256 pad: 1 kernel_size: 3 } }
layer { name: "relu3_1" type: "ReLU" bottom: "conv3_1" top: "conv3_1" }
layer { name: "conv3_2" type: "Convolution" bottom: "conv3_1" top: "conv3_2"
  convolution_param { num_output: 256 pad: 1 kernel_size: 3 } }
layer { name: "relu3_2" type: "ReLU" bottom: "conv3_2" top: "conv3_2" }
layer { name: "conv3_3" type: "Convolution" bottom: "conv3_2" top: "conv3_3"
  convolution_param { num_output: 256 pad: 1 kernel_size: 3 } }
layer { name: "relu3_3" type: "ReLU" bottom: "conv3_3" top: "conv3_3" }
layer { name: "pool3" type: "Pooling" bottom: "conv3_3" top: "pool3"
  pooling_param { pool: MAX kernel_size: 2 stride: 2 } }

layer { name: "conv4_1" type: "Convolution" bottom: "pool3" top: "conv4_1"
  convolution_param { num_output: 512 pad: 1 kernel_size: 3 } }
layer { name: "relu4_1" type: "ReLU" bottom: "conv4_1" top: "conv4_1" }
layer { name: "conv4_2" type: "Convolution" bottom: "conv4_1" top: "conv4_2"
  convolution_param { num_output: 512 pad: 1 kernel_size: 3 } }
layer { name: "relu4_2" type: "ReLU" bottom: "conv4_2" top: "conv4_2" }
layer { name: "conv4_3" type: "Convolution" bottom: "conv4_2" top: "conv4_3"
  convolution_param { num_output: 512 pad: 1 kernel_size: 3 } }
layer { name: "relu4_3" type: "ReLU" bottom: "conv4_3" top: "conv4_3" }
layer { name: "pool4" type: "Pooling" bottom: "conv4_3" top: "pool4"
  pooling_param { pool: MAX kernel_size: 2 stride: 1 } }

layer { name: "conv5_1" type: "Convolution" bottom: "pool4" top: "conv5_1"
  convolution_param { num_output: 512 pad: 2 kernel_size: 3 dilation: 2 } }
layer { name: "relu5_1" type: "ReLU" bottom: "conv5_1" top: "conv5_1" }
layer { name: "conv5_2" type: "Convolution" bottom: "conv5_1" top: "conv5_2"
  convolution_param { num_output: 512 pad: 2 kernel_size: 3 dilation: 2 } }
layer { name: "relu5_2" type: "ReLU" bottom: "conv5_2" top: "conv5_2" }
layer { name: "conv5_3" type: "Convolution" bottom: "conv5_2" top: "conv5_3"
  convolution_param { num_output: 512 pad: 2 kernel_size: 3 dilation: 2 } }
layer { name: "relu5_3" type: "ReLU" bottom: "conv5_3" top: "conv5_3" }

layer { name: "conv1_1_down" type: "Convolution" bottom: "conv1_1" top: "conv1_1_down"
  convolution_param { num_output: 21 kernel_size: 1 } }
layer { name: "conv1_2_down" type: "Convolution" bottom: "conv1_2" top: "conv1_2_down"
  convolution_param { num_output: 21 kernel_size: 1 } }
layer { name: "score_fuse1" type: "Eltwise" bottom: "conv1_1_down" bottom: "conv1_2_down"
  top: "score_fuse1" eltwise_param { operation: SUM } }
layer { name: "score-dsn1" type: "Convolution" bottom: "score_fuse1" top: "upscore-dsn1"
  convolution_param { num_output: 1 kernel_size: 1 } }
layer { name: "sigmoid-dsn1" type: "Sigmoid" bottom: "upscore-dsn1" top: "sigmoid-dsn1"}

layer { name: "conv2_1_down" type: "Convolution" bottom: "conv2_1" top: "conv2_1_down"
  convolution_param { num_output: 21 kernel_size: 1 } }
layer { name: "conv2_2_down" type: "Convolution" bottom: "conv2_2" top: "conv2_2_down"
  convolution_param { num_output: 21 kernel_size: 1 } }
layer { name: "score_fuse2" type: "Eltwise" bottom: "conv2_1_down" bottom: "conv2_2_down"
  top: "score_fuse2" eltwise_param { operation: SUM } }
layer { name: "score-dsn2" type: "Convolution" bottom: "score_fuse2" top: "score-dsn2"
  convolution_param { num_output: 1 kernel_size: 1 } }
layer { name: "upsample_2" type: "Deconvolution" bottom: "score-dsn2" top: "score-dsn2-up"
  convolution_param { kernel_size: 4 stride: 2 num_output: 1 } }
layer { name: "crop2" type: "Crop" bottom: "score-dsn2-up" bottom: "data" top: "upscore-dsn2"
  crop_param { axis: 2 offset: 1 } }
layer { name: "sigmoid-dsn2" type: "Sigmoid" bottom: "upscore-dsn2" top: "sigmoid-dsn2"}

layer { name: "conv3_1_down" type: "Convolution" bottom: "conv3_1" top: "conv3_1_down"
  convolution_param { num_output: 21 kernel_size: 1 } }
layer { name: "conv3_2_down" type: "Convolution" bottom: "conv3_2" top: "conv3_2_down"
  convolution_param { num_output: 21 kernel_size: 1 } }
layer { name: "conv3_3_down" type: "Convolution" bottom: "conv3_3" top: "conv3_3_down"
  convolution_param { num_output: 21 kernel_size: 1 } }
layer { name: "score_fuse3" type: "Eltwise" bottom: "conv3_1_down" bottom: "conv3_2_down"
  bottom: "conv3_3_down" top: "score_fuse3" eltwise_param { operation: SUM } }
layer { name: "score-dsn3" type: "Convolution" bottom: "score_fuse3" top: "score-dsn3"
  convolution_param { num_output: 1 kernel_size: 1 } }
layer { name: "upsample_4" type: "Deconvolution" bottom: "score-dsn3" top: "score-dsn3-up"
  convolution_param { kernel_size: 8 stride: 4 num_output: 1 } }
layer { name: "crop3" type: "Crop" bottom: "score-dsn3-up" bottom: "data" top: "upscore-dsn3"
  crop_param { axis: 2 offset: 2 } }
layer { name: "sigmoid-dsn3" type: "Sigmoid" bottom: "upscore-dsn3" top: "sigmoid-dsn3"}

layer { name: "conv4_1_down" type: "Convolution" bottom: "conv4_1" top: "conv4_1_down"
  convolution_param { num_output: 21 kernel_size: 1 } }
layer { name: "conv4_2_down" type: "Convolution" bottom: "conv4_2" top: "conv4_2_down"
  convolution_param { num_output: 21 kernel_size: 1 } }
layer { name: "conv4_3_down" type: "Convolution" bottom: "conv4_3" top: "conv4_3_down"
  convolution_param { num_output: 21 kernel_size: 1 } }
layer { name: "score_fuse4" type: "Eltwise" bottom: "conv4_1_down" bottom: "conv4_2_down"
  bottom: "conv4_3_down" top: "score_fuse4" eltwise_param { operation: SUM } }
layer { name: "score-dsn4" type: "Convolution" bottom: "score_fuse4" top: "score-dsn4"
  convolution_param { num_output: 1 kernel_size: 1 } }
layer { name: "upsample_8" type: "Deconvolution" bottom: "score-dsn4" top: "score-dsn4-up"
  convolution_param { kernel_size: 16 stride: 8 num_output: 1 } }
layer { name: "crop4" type: "Crop" bottom: "score-dsn4-up" bottom: "data" top: "upscore-dsn4"
  crop_param { axis: 2 offset: 4 } }
layer { name: "sigmoid-dsn4" type: "Sigmoid" bottom: "upscore-dsn4" top: "sigmoid-dsn4"}

layer { name: "conv5_1_down" type: "Convolution" bottom: "conv5_1" top: "conv5_1_down"
  convolution_param { num_output: 21 kernel_size: 1 } }
layer { name: "conv5_2_down" type: "Convolution" bottom: "conv5_2" top: "conv5_2_down"
  convolution_param { num_output: 21 kernel_size: 1 } }
layer { name: "conv5_3_down" type: "Convolution" bottom: "conv5_3" top: "conv5_3_down"
  convolution_param { num_output: 21 kernel_size: 1 } }
layer { name: "score_fuse5" type: "Eltwise" bottom: "conv5_1_down" bottom: "conv5_2_down"
  bottom: "conv5_3_down" top: "score_fuse5" eltwise_param { operation: SUM } }
layer { name: "score-dsn5" type: "Convolution" bottom: "score_fuse5" top: "score-dsn5"
  convolution_param { num_output: 1 kernel_size: 1 } }
layer { name: "upsample_8_5" type: "Deconvolution" bottom: "score-dsn5" top: "score-dsn5-up"
  convolution_param { kernel_size: 16 stride: 8 num_output: 1 } }
layer { name: "crop5" type: "Crop" bottom: "score-dsn5-up" bottom: "data" top: "upscore-dsn5"
  crop_param { axis: 2 offset: 0 } }
layer { name: "sigmoid-dsn5" type: "Sigmoid" bottom: "upscore-dsn5" top: "sigmoid-dsn5"}

layer { name: "concat" bottom: "upscore-dsn1"  bottom: "upscore-dsn2" bottom: "upscore-dsn3"
  bottom: "upscore-dsn4" bottom: "upscore-dsn5" top: "concat-upscore" type: "Concat"
  concat_param { concat_dim: 1 } }
layer { name: "new-score-weighting" type: "Convolution" bottom: "concat-upscore" top: "upscore-fuse"
  convolution_param { num_output: 1 kernel_size: 1 } }
layer { name: "sigmoid-fuse" type: "Sigmoid" bottom: "upscore-fuse" top: "sigmoid-fuse"}
athiselvam commented 6 years ago

@yun-liu Thanks a lot, man. It's working perfectly.

piaobuliao commented 5 years ago

@yun-liu 224*224的图像的crop这样配置 还是任意图像的都这样配置？

yun-liu commented 5 years ago

@piaobuliao 任意的