Created
November 22, 2017 05:13
-
-
Save dongzhuoyao/e532cdda399b86fdfc6af181cf89bb30 to your computer and use it in GitHub Desktop.
segaware_train_voc_train_aug
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| # VGG 16-layer network convolutional finetuning | |
| # Network modified to have smaller receptive field (128 pixels) | |
| # and smaller stride (8 pixels) when run in convolutional mode. | |
| # | |
| # In this model we also change max pooling size in the first 4 layer | |
| # from 2 to 3 while retaining stride = 2 | |
| # which makes it easier to exactly align responses at different layer. | |
| # | |
| # For alignment to work, we set (we choose 32x so as to be able to evaluate | |
| # the model for all different subsampling sizes): | |
| # (1) input dimension equal to | |
| # $n = 32 * k - 31$, e.g., 321 (for k = 11) | |
| # Dimension after pooling w. subsampling: | |
| # (16 * k - 15); (8 * k - 7); (4 * k - 3); (2 * k - 1); (k). | |
| # For k = 11, these translate to | |
| # 161; 81; 41; 21; 11 | |
| # | |
| name: "segaware-all-largeFOV" | |
| layer { | |
| name: "data" | |
| type: "TwoImageData" | |
| top: "data" | |
| top: "label" | |
| image_data_param { | |
| root_folder: "/opt/home/aharley/datasets/VOC2012" | |
| source: "segaware/list/voc_train_aug.txt" | |
| batch_size: 1 | |
| shuffle: true | |
| } | |
| two_image_data_param { | |
| first_is_color: true | |
| second_is_color: false | |
| } | |
| transform_param { | |
| mean_value: 104.008 | |
| mean_value: 116.669 | |
| mean_value: 122.675 | |
| crop_size: 321 | |
| mirror: true | |
| } | |
| } | |
| # Embeddings | |
| # rgb for emb | |
| layer { | |
| bottom: "data" | |
| name: "rgb_emb" | |
| top: "rgb_emb" | |
| type: "Power" | |
| power_param { | |
| scale: 0.0039215 # 1/255 | |
| } | |
| } | |
| # Embedding layers | |
| layer { | |
| bottom: "data" | |
| top: "emb1_1" | |
| name: "emb1_1" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 64 | |
| kernel_size: 3 pad: 1 | |
| stride: 1 | |
| weight_filler { type: "xavier" } | |
| bias_filler { type: "constant" } | |
| } | |
| } | |
| layer { | |
| bottom: "emb1_1" | |
| top: "emb1_1" | |
| name: "relu1_1" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "emb1_1" | |
| top: "emb1_2" | |
| name: "emb1_2" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 64 | |
| kernel_size: 3 pad: 1 | |
| weight_filler { type: "xavier" } | |
| bias_filler { type: "constant" } | |
| } | |
| } | |
| layer { | |
| bottom: "emb1_2" | |
| top: "emb1_2" | |
| name: "relu1_2" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "emb1_2" | |
| top: "emb_pool1" | |
| name: "emb_pool1" | |
| type: "Pooling" | |
| pooling_param { | |
| pool: MAX | |
| kernel_size: 3 pad: 1 | |
| stride: 2 | |
| } | |
| } | |
| layer { | |
| bottom: "emb_pool1" | |
| top: "emb2_1" | |
| name: "emb2_1" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 128 | |
| kernel_size: 3 pad: 1 | |
| weight_filler { type: "xavier" } | |
| bias_filler { type: "constant" } | |
| } | |
| } | |
| layer { | |
| bottom: "emb2_1" | |
| top: "emb2_1" | |
| name: "relu2_1" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "emb2_1" | |
| top: "emb2_2" | |
| name: "emb2_2" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 128 | |
| kernel_size: 3 pad: 1 | |
| weight_filler { type: "xavier" } | |
| bias_filler { type: "constant" } | |
| } | |
| } | |
| layer { | |
| bottom: "emb2_2" | |
| top: "emb2_2" | |
| name: "relu2_2" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "emb2_2" | |
| top: "emb_pool2" | |
| name: "emb_pool2" | |
| type: "Pooling" | |
| pooling_param { | |
| pool: MAX | |
| kernel_size: 3 pad: 1 | |
| stride: 2 | |
| } | |
| } | |
| layer { | |
| bottom: "emb_pool2" | |
| top: "emb3_1" | |
| name: "emb3_1" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 256 | |
| kernel_size: 3 pad: 1 | |
| weight_filler { type: "xavier" } | |
| bias_filler { type: "constant" } | |
| } | |
| } | |
| layer { | |
| bottom: "emb3_1" | |
| top: "emb3_1" | |
| name: "relu3_1" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "emb3_1" | |
| top: "emb3_2" | |
| name: "emb3_2" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 256 | |
| kernel_size: 3 pad: 1 | |
| weight_filler { type: "xavier" } | |
| bias_filler { type: "constant" } | |
| } | |
| } | |
| layer { | |
| bottom: "emb3_2" | |
| top: "emb3_2" | |
| name: "relu3_2" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "emb3_2" | |
| top: "emb3_3" | |
| name: "emb3_3" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 256 | |
| kernel_size: 3 pad: 1 | |
| weight_filler { type: "xavier" } | |
| bias_filler { type: "constant" } | |
| } | |
| } | |
| layer { | |
| bottom: "emb3_3" | |
| top: "emb3_3" | |
| name: "relu3_3" | |
| type: "ReLU" | |
| } | |
| # resize, for uniformity | |
| layer { | |
| type: "Interp" | |
| bottom: "emb2_2" | |
| name: "emb22_resized" | |
| top: "emb22_resized" | |
| interp_param { | |
| height: 321 | |
| width: 321 | |
| } | |
| } | |
| layer { | |
| type: "Interp" | |
| bottom: "emb3_3" | |
| name: "emb33_resized" | |
| top: "emb33_resized" | |
| interp_param { | |
| height: 321 | |
| width: 321 | |
| } | |
| } | |
| # Concat, and weight | |
| layer { | |
| bottom: "rgb_emb" | |
| bottom: "emb1_2" | |
| bottom: "emb22_resized" | |
| bottom: "emb33_resized" | |
| top: "fused" | |
| name: "fused" | |
| type: "Concat" | |
| } | |
| layer { | |
| bottom: "fused" | |
| top: "emb" | |
| name: "emb" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 64 | |
| kernel_size: 1 pad: 0 | |
| weight_filler { type: "xavier" } | |
| bias_filler { type: "constant" } | |
| } | |
| } | |
| ###### Mask prep for conv1 | |
| layer { | |
| bottom: "emb" | |
| top: "emb_dist1" | |
| name: "emb_dist1" | |
| type: "Im2dist" | |
| convolution_param { | |
| kernel_size: 3 pad: 1 | |
| stride: 1 | |
| } | |
| im2dist_param { norm: L1 } | |
| } | |
| layer { | |
| bottom: "emb_dist1" | |
| top: "weighted_emb_dist1_1" | |
| name: "weighted_emb_dist1_1" | |
| type: "Scale" | |
| param { lr_mult: 1 decay_mult: 1 } | |
| scale_param { num_axes: 0 filler { type: "constant" value: 0.05 } bias_term: false } | |
| } | |
| layer { bottom: "weighted_emb_dist1_1" top: "abs_emb_dist1_1" name: "abs_emb_dist1_1" type: "AbsVal" } | |
| layer { bottom: "abs_emb_dist1_1" top: "exp_emb_dist1_1" name: "exp_emb_dist1_1" type: "Exp" exp_param { scale: -1 } } | |
| layer { bottom: "exp_emb_dist1_1" top: "exp_emb_dist1_1_3chan" name: "exp_emb_dist1_1_3chan" type: "Tile" tile_param { axis: 1 tiles: 3 }} | |
| layer { | |
| bottom: "emb_dist1" | |
| top: "weighted_emb_dist1_2" | |
| name: "weighted_emb_dist1_2" | |
| type: "Scale" | |
| param { lr_mult: 1 decay_mult: 1 } | |
| scale_param { num_axes: 0 filler { type: "constant" value: 0.05 } bias_term: false } | |
| } | |
| layer { bottom: "weighted_emb_dist1_2" top: "abs_emb_dist1_2" name: "abs_emb_dist1_2" type: "AbsVal"} | |
| layer { bottom: "abs_emb_dist1_2" top: "exp_emb_dist1_2" name: "exp_emb_dist1_2" type: "Exp" exp_param { scale: -1 }} | |
| layer { bottom: "exp_emb_dist1_2" top: "exp_emb_dist1_2_64chan" name: "exp_emb_dist1_2_64chan" type: "Tile" tile_param { axis: 1 tiles: 64 }} | |
| layer { | |
| bottom: "data" | |
| top: "im2col_conv1_1" | |
| name: "im2col_conv1_1" | |
| type: "Im2col" | |
| convolution_param { | |
| num_output: 64 kernel_size: 3 pad: 1 | |
| } | |
| } | |
| layer { | |
| bottom: "im2col_conv1_1" | |
| bottom: "exp_emb_dist1_1_3chan" | |
| top: "im2col_conv1_1_hit" | |
| name: "im2col_conv1_1_hit" | |
| type: "Eltwise" | |
| eltwise_param { operation: PROD } | |
| } | |
| layer { | |
| bottom: "im2col_conv1_1_hit" | |
| top: "conv1_1" | |
| name: "conv1_1" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 64 kernel_size: 3 pad: 1 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "conv1_1" | |
| top: "conv1_1" | |
| name: "relu1_1" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "conv1_1" | |
| top: "im2col_conv1_2" | |
| name: "im2col_conv1_2" | |
| type: "Im2col" | |
| convolution_param { | |
| num_output: 64 kernel_size: 3 pad: 1 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "im2col_conv1_2" | |
| bottom: "exp_emb_dist1_2_64chan" | |
| top: "im2col_conv1_2_hit" | |
| name: "im2col_conv1_2_hit" | |
| type: "Eltwise" | |
| eltwise_param { operation: PROD } | |
| } | |
| layer { | |
| bottom: "im2col_conv1_2_hit" | |
| top: "conv1_2" | |
| name: "conv1_2" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 64 kernel_size: 3 pad: 1 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "conv1_2" | |
| top: "conv1_2" | |
| name: "relu1_2" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "conv1_2" | |
| top: "pool1" | |
| name: "pool1" | |
| type: "Pooling" | |
| pooling_param { | |
| pool: MAX | |
| kernel_size: 3 pad: 1 stride: 2 | |
| } | |
| } | |
| ###### Mask prep for conv2 | |
| layer { | |
| bottom: "emb" | |
| top: "emb_shrink2" | |
| name: "emb_shrink2" | |
| type: "Interp" | |
| interp_param { | |
| shrink_factor: 2 | |
| pad_beg: 0 | |
| pad_end: 0 | |
| } | |
| } | |
| layer { | |
| bottom: "emb_shrink2" | |
| top: "emb_dist2" | |
| name: "emb_dist2" | |
| type: "Im2dist" | |
| convolution_param { | |
| kernel_size: 3 pad: 1 | |
| stride: 1 | |
| } | |
| im2dist_param { norm: L1 } | |
| } | |
| layer { | |
| bottom: "emb_dist2" | |
| top: "weighted_emb_dist2_1" | |
| name: "weighted_emb_dist2_1" | |
| type: "Scale" | |
| param { lr_mult: 1 decay_mult: 1 } | |
| scale_param { num_axes: 0 filler { type: "constant" value: 0.05 } bias_term: false } | |
| } | |
| layer { bottom: "weighted_emb_dist2_1" top: "abs_emb_dist2_1" name: "abs_emb_dist2_1" type: "AbsVal"} | |
| layer { bottom: "abs_emb_dist2_1" top: "exp_emb_dist2_1" name: "exp_emb_dist2_1" type: "Exp" exp_param { scale: -1 }} | |
| layer { bottom: "exp_emb_dist2_1" top: "exp_emb_dist2_1_64chan" name: "exp_emb_dist2_1_64chan" type: "Tile" tile_param { axis: 1 tiles: 64 }} | |
| layer { | |
| bottom: "emb_dist2" | |
| top: "weighted_emb_dist2_2" | |
| name: "weighted_emb_dist2_2" | |
| type: "Scale" | |
| param { lr_mult: 1 decay_mult: 1 } | |
| scale_param { num_axes: 0 filler { type: "constant" value: 0.05 } bias_term: false } | |
| } | |
| layer { bottom: "weighted_emb_dist2_2" top: "abs_emb_dist2_2" name: "abs_emb_dist2_2" type: "AbsVal"} | |
| layer { bottom: "abs_emb_dist2_2" top: "exp_emb_dist2_2" name: "exp_emb_dist2_2" type: "Exp" exp_param { scale: -1 }} | |
| layer { bottom: "exp_emb_dist2_2" top: "exp_emb_dist2_2_128chan" name: "exp_emb_dist2_2_128chan" type: "Tile" tile_param { axis: 1 tiles: 128 }} | |
| layer { | |
| bottom: "pool1" | |
| top: "im2col_conv2_1" | |
| name: "im2col_conv2_1" | |
| type: "Im2col" | |
| convolution_param { | |
| num_output: 128 kernel_size: 3 pad: 1 | |
| } | |
| } | |
| layer { | |
| bottom: "im2col_conv2_1" | |
| bottom: "exp_emb_dist2_1_64chan" | |
| top: "im2col_conv2_1_hit" | |
| name: "im2col_conv2_1_hit" | |
| type: "Eltwise" | |
| eltwise_param { operation: PROD } | |
| } | |
| layer { | |
| bottom: "im2col_conv2_1_hit" | |
| top: "conv2_1" | |
| name: "conv2_1" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 128 kernel_size: 3 pad: 1 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "conv2_1" | |
| top: "conv2_1" | |
| name: "relu2_1" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "conv2_1" | |
| top: "im2col_conv2_2" | |
| name: "im2col_conv2_2" | |
| type: "Im2col" | |
| convolution_param { | |
| num_output: 128 kernel_size: 3 pad: 1 | |
| } | |
| } | |
| layer { | |
| bottom: "im2col_conv2_2" | |
| bottom: "exp_emb_dist2_2_128chan" | |
| top: "im2col_conv2_2_hit" | |
| name: "im2col_conv2_2_hit" | |
| type: "Eltwise" | |
| eltwise_param { operation: PROD } | |
| } | |
| layer { | |
| bottom: "im2col_conv2_2_hit" | |
| top: "conv2_2" | |
| name: "conv2_2" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 128 kernel_size: 3 pad: 1 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "conv2_2" | |
| top: "conv2_2" | |
| name: "relu2_2" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "conv2_2" | |
| top: "pool2" | |
| name: "pool2" | |
| type: "Pooling" | |
| pooling_param { | |
| pool: MAX | |
| kernel_size: 3 pad: 1 stride: 2 | |
| } | |
| } | |
| ###### Mask prep for conv3 | |
| layer { | |
| bottom: "emb" | |
| top: "emb_shrink4" | |
| name: "emb_shrink4" | |
| type: "Interp" | |
| interp_param { | |
| shrink_factor: 4 | |
| pad_beg: 0 | |
| pad_end: 0 | |
| } | |
| } | |
| layer { | |
| bottom: "emb_shrink4" | |
| top: "emb_dist3" | |
| name: "emb_dist3" | |
| type: "Im2dist" | |
| convolution_param { | |
| kernel_size: 3 pad: 1 | |
| stride: 1 | |
| } | |
| im2dist_param { norm: L1 } | |
| } | |
| layer { | |
| bottom: "emb_dist3" | |
| top: "weighted_emb_dist3_1" | |
| name: "weighted_emb_dist3_1" | |
| type: "Scale" | |
| param { lr_mult: 1 decay_mult: 1 } | |
| scale_param { num_axes: 0 filler { type: "constant" value: 0.05 } bias_term: false } | |
| } | |
| layer { bottom: "weighted_emb_dist3_1" top: "abs_emb_dist3_1" name: "abs_emb_dist3_1" type: "AbsVal"} | |
| layer { bottom: "abs_emb_dist3_1" top: "exp_emb_dist3_1" name: "exp_emb_dist3_1" type: "Exp" exp_param { scale: -1 }} | |
| layer { bottom: "exp_emb_dist3_1" top: "exp_emb_dist3_1_128chan" name: "exp_emb_dist3_1_128chan" type: "Tile" tile_param { axis: 1 tiles: 128 }} | |
| layer { | |
| bottom: "emb_dist3" | |
| top: "weighted_emb_dist3_2" | |
| name: "weighted_emb_dist3_2" | |
| type: "Scale" | |
| param { lr_mult: 1 decay_mult: 1 } | |
| scale_param { num_axes: 0 filler { type: "constant" value: 0.05 } bias_term: false } | |
| } | |
| layer { bottom: "weighted_emb_dist3_2" top: "abs_emb_dist3_2" name: "abs_emb_dist3_2" type: "AbsVal"} | |
| layer { bottom: "abs_emb_dist3_2" top: "exp_emb_dist3_2" name: "exp_emb_dist3_2" type: "Exp" exp_param { scale: -1 }} | |
| layer { bottom: "exp_emb_dist3_2" top: "exp_emb_dist3_2_256chan" name: "exp_emb_dist3_2_256chan" type: "Tile" tile_param { axis: 1 tiles: 256 }} | |
| layer { | |
| bottom: "emb_dist3" | |
| top: "weighted_emb_dist3_3" | |
| name: "weighted_emb_dist3_3" | |
| type: "Scale" | |
| param { lr_mult: 1 decay_mult: 1 } | |
| scale_param { num_axes: 0 filler { type: "constant" value: 0.05 } bias_term: false } | |
| } | |
| layer { bottom: "weighted_emb_dist3_3" top: "abs_emb_dist3_3" name: "abs_emb_dist3_3" type: "AbsVal"} | |
| layer { bottom: "abs_emb_dist3_3" top: "exp_emb_dist3_3" name: "exp_emb_dist3_3" type: "Exp" exp_param { scale: -1 }} | |
| layer { bottom: "exp_emb_dist3_3" top: "exp_emb_dist3_3_256chan" name: "exp_emb_dist3_3_256chan" type: "Tile" tile_param { axis: 1 tiles: 256 }} | |
| layer { | |
| bottom: "pool2" | |
| top: "im2col_conv3_1" | |
| name: "im2col_conv3_1" | |
| type: "Im2col" | |
| convolution_param { | |
| num_output: 128 kernel_size: 3 pad: 1 | |
| } | |
| } | |
| layer { | |
| bottom: "im2col_conv3_1" | |
| bottom: "exp_emb_dist3_1_128chan" | |
| top: "im2col_conv3_1_hit" | |
| name: "im2col_conv3_1_hit" | |
| type: "Eltwise" | |
| eltwise_param { operation: PROD } | |
| } | |
| layer { | |
| bottom: "im2col_conv3_1_hit" | |
| top: "conv3_1" | |
| name: "conv3_1" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 256 kernel_size: 3 pad: 1 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "conv3_1" | |
| top: "conv3_1" | |
| name: "relu3_1" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "conv3_1" | |
| top: "im2col_conv3_2" | |
| name: "im2col_conv3_2" | |
| type: "Im2col" | |
| convolution_param { | |
| num_output: 256 kernel_size: 3 pad: 1 | |
| } | |
| } | |
| layer { | |
| bottom: "im2col_conv3_2" | |
| bottom: "exp_emb_dist3_2_256chan" | |
| top: "im2col_conv3_2_hit" | |
| name: "im2col_conv3_2_hit" | |
| type: "Eltwise" | |
| eltwise_param { operation: PROD } | |
| } | |
| layer { | |
| bottom: "im2col_conv3_2_hit" | |
| top: "conv3_2" | |
| name: "conv3_2" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 256 kernel_size: 3 pad: 1 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "conv3_2" | |
| top: "conv3_2" | |
| name: "relu3_2" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "conv3_2" | |
| top: "im2col_conv3_3" | |
| name: "im2col_conv3_3" | |
| type: "Im2col" | |
| convolution_param { | |
| num_output: 256 kernel_size: 3 pad: 1 | |
| } | |
| } | |
| layer { | |
| bottom: "im2col_conv3_3" | |
| bottom: "exp_emb_dist3_3_256chan" | |
| top: "im2col_conv3_3_hit" | |
| name: "im2col_conv3_3_hit" | |
| type: "Eltwise" | |
| eltwise_param { operation: PROD } | |
| } | |
| layer { | |
| bottom: "im2col_conv3_3_hit" | |
| top: "conv3_3" | |
| name: "conv3_3" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 256 kernel_size: 3 pad: 1 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "conv3_3" | |
| top: "conv3_3" | |
| name: "relu3_3" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "conv3_3" | |
| top: "pool3" | |
| name: "pool3" | |
| type: "Pooling" | |
| pooling_param { | |
| pool: MAX | |
| kernel_size: 3 pad: 1 | |
| stride: 2 | |
| } | |
| } | |
| ###### Mask prep for conv4 | |
| layer { | |
| bottom: "emb" | |
| top: "emb_shrink8" | |
| name: "emb_shrink8" | |
| type: "Interp" | |
| interp_param { | |
| shrink_factor: 8 | |
| pad_beg: 0 | |
| pad_end: 0 | |
| } | |
| } | |
| layer { | |
| bottom: "emb_shrink8" | |
| top: "emb_dist4" | |
| name: "emb_dist4" | |
| type: "Im2dist" | |
| convolution_param { | |
| kernel_size: 3 pad: 1 | |
| stride: 1 | |
| } | |
| im2dist_param { norm: L1 } | |
| } | |
| layer { | |
| bottom: "emb_dist4" | |
| top: "weighted_emb_dist4_1" | |
| name: "weighted_emb_dist4_1" | |
| type: "Scale" | |
| param { lr_mult: 1 decay_mult: 1 } | |
| scale_param { num_axes: 0 filler { type: "constant" value: 0.05 } bias_term: false } | |
| } | |
| layer { bottom: "weighted_emb_dist4_1" top: "abs_emb_dist4_1" name: "abs_emb_dist4_1" type: "AbsVal"} | |
| layer { bottom: "abs_emb_dist4_1" top: "exp_emb_dist4_1" name: "exp_emb_dist4_1" type: "Exp" exp_param { scale: -1 }} | |
| layer { bottom: "exp_emb_dist4_1" top: "exp_emb_dist4_1_256chan" name: "exp_emb_dist4_1_256chan" type: "Tile" tile_param { axis: 1 tiles: 256 }} | |
| layer { | |
| bottom: "emb_dist4" | |
| top: "weighted_emb_dist4_2" | |
| name: "weighted_emb_dist4_2" | |
| type: "Scale" | |
| param { lr_mult: 1 decay_mult: 1 } | |
| scale_param { num_axes: 0 filler { type: "constant" value: 0.05 } bias_term: false } | |
| } | |
| layer { bottom: "weighted_emb_dist4_2" top: "abs_emb_dist4_2" name: "abs_emb_dist4_2" type: "AbsVal"} | |
| layer { bottom: "abs_emb_dist4_2" top: "exp_emb_dist4_2" name: "exp_emb_dist4_2" type: "Exp" exp_param { scale: -1 }} | |
| layer { bottom: "exp_emb_dist4_2" top: "exp_emb_dist4_2_512chan" name: "exp_emb_dist4_2_512chan" type: "Tile" tile_param { axis: 1 tiles: 512 }} | |
| layer { | |
| bottom: "emb_dist4" | |
| top: "weighted_emb_dist4_3" | |
| name: "weighted_emb_dist4_3" | |
| type: "Scale" | |
| param { lr_mult: 1 decay_mult: 1 } | |
| scale_param { num_axes: 0 filler { type: "constant" value: 0.05 } bias_term: false } | |
| } | |
| layer { bottom: "weighted_emb_dist4_3" top: "abs_emb_dist4_3" name: "abs_emb_dist4_3" type: "AbsVal"} | |
| layer { bottom: "abs_emb_dist4_3" top: "exp_emb_dist4_3" name: "exp_emb_dist4_3" type: "Exp" exp_param { scale: -1 }} | |
| layer { bottom: "exp_emb_dist4_3" top: "exp_emb_dist4_3_512chan" name: "exp_emb_dist4_3_512chan" type: "Tile" tile_param { axis: 1 tiles: 512 }} | |
| layer { | |
| bottom: "pool3" | |
| top: "im2col_conv4_1" | |
| name: "im2col_conv4_1" | |
| type: "Im2col" | |
| convolution_param { | |
| num_output: 512 kernel_size: 3 pad: 1 | |
| } | |
| } | |
| layer { | |
| bottom: "im2col_conv4_1" | |
| bottom: "exp_emb_dist4_1_256chan" | |
| top: "im2col_conv4_1_hit" | |
| name: "im2col_conv4_1_hit" | |
| type: "Eltwise" | |
| eltwise_param { operation: PROD } | |
| } | |
| layer { | |
| bottom: "im2col_conv4_1_hit" | |
| top: "conv4_1" | |
| name: "conv4_1" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 512 kernel_size: 3 pad: 1 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "conv4_1" | |
| top: "conv4_1" | |
| name: "relu4_1" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "conv4_1" | |
| top: "im2col_conv4_2" | |
| name: "im2col_conv4_2" | |
| type: "Im2col" | |
| convolution_param { | |
| num_output: 512 kernel_size: 3 pad: 1 | |
| } | |
| } | |
| layer { | |
| bottom: "im2col_conv4_2" | |
| bottom: "exp_emb_dist4_2_512chan" | |
| top: "im2col_conv4_2_hit" | |
| name: "im2col_conv4_2_hit" | |
| type: "Eltwise" | |
| eltwise_param { operation: PROD } | |
| } | |
| layer { | |
| bottom: "im2col_conv4_2_hit" | |
| top: "conv4_2" | |
| name: "conv4_2" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 512 kernel_size: 3 pad: 1 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "conv4_2" | |
| top: "conv4_2" | |
| name: "relu4_2" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "conv4_2" | |
| top: "im2col_conv4_3" | |
| name: "im2col_conv4_3" | |
| type: "Im2col" | |
| convolution_param { | |
| num_output: 512 kernel_size: 3 pad: 1 | |
| } | |
| } | |
| layer { | |
| bottom: "im2col_conv4_3" | |
| bottom: "exp_emb_dist4_3_512chan" | |
| top: "im2col_conv4_3_hit" | |
| name: "im2col_conv4_3_hit" | |
| type: "Eltwise" | |
| eltwise_param { operation: PROD } | |
| } | |
| layer { | |
| bottom: "im2col_conv4_3_hit" | |
| top: "conv4_3" | |
| name: "conv4_3" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 512 kernel_size: 3 pad: 1 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "conv4_3" | |
| top: "conv4_3" | |
| name: "relu4_3" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "conv4_3" | |
| top: "pool4" | |
| name: "pool4" | |
| type: "Pooling" | |
| pooling_param { | |
| pool: MAX | |
| kernel_size: 3 pad: 1 stride: 1 | |
| } | |
| } | |
| ###### Mask prep for conv5 | |
| layer { | |
| bottom: "emb_shrink8" | |
| top: "emb_dist5" | |
| name: "emb_dist5" | |
| type: "Im2dist" | |
| convolution_param { | |
| kernel_size: 3 pad: 2 dilation: 2 | |
| stride: 1 | |
| } | |
| im2dist_param { norm: L1 } | |
| } | |
| layer { | |
| bottom: "emb_dist5" | |
| top: "weighted_emb_dist5_1" | |
| name: "weighted_emb_dist5_1" | |
| type: "Scale" | |
| param { lr_mult: 1 decay_mult: 1 } | |
| scale_param { num_axes: 0 filler { type: "constant" value: 0.05 } bias_term: false } | |
| } | |
| layer { bottom: "weighted_emb_dist5_1" top: "abs_emb_dist5_1" name: "abs_emb_dist5_1" type: "AbsVal"} | |
| layer { bottom: "abs_emb_dist5_1" top: "exp_emb_dist5_1" name: "exp_emb_dist5_1" type: "Exp" exp_param { scale: -1 }} | |
| layer { bottom: "exp_emb_dist5_1" top: "exp_emb_dist5_1_512chan" name: "exp_emb_dist5_1_512chan" type: "Tile" tile_param { axis: 1 tiles: 512 }} | |
| layer { | |
| bottom: "emb_dist5" | |
| top: "weighted_emb_dist5_2" | |
| name: "weighted_emb_dist5_2" | |
| type: "Scale" | |
| param { lr_mult: 1 decay_mult: 1 } | |
| scale_param { num_axes: 0 filler { type: "constant" value: 0.05 } bias_term: false } | |
| } | |
| layer { bottom: "weighted_emb_dist5_2" top: "abs_emb_dist5_2" name: "abs_emb_dist5_2" type: "AbsVal"} | |
| layer { bottom: "abs_emb_dist5_2" top: "exp_emb_dist5_2" name: "exp_emb_dist5_2" type: "Exp" exp_param { scale: -1 }} | |
| layer { bottom: "exp_emb_dist5_2" top: "exp_emb_dist5_2_512chan" name: "exp_emb_dist5_2_512chan" type: "Tile" tile_param { axis: 1 tiles: 512 }} | |
| layer { | |
| bottom: "emb_dist5" | |
| top: "weighted_emb_dist5_3" | |
| name: "weighted_emb_dist5_3" | |
| type: "Scale" | |
| param { lr_mult: 1 decay_mult: 1 } | |
| scale_param { num_axes: 0 filler { type: "constant" value: 0.05 } bias_term: false } | |
| } | |
| layer { bottom: "weighted_emb_dist5_3" top: "abs_emb_dist5_3" name: "abs_emb_dist5_3" type: "AbsVal"} | |
| layer { bottom: "abs_emb_dist5_3" top: "exp_emb_dist5_3" name: "exp_emb_dist5_3" type: "Exp" exp_param { scale: -1 }} | |
| layer { bottom: "exp_emb_dist5_3" top: "exp_emb_dist5_3_512chan" name: "exp_emb_dist5_3_512chan" type: "Tile" tile_param { axis: 1 tiles: 512 }} | |
| ## conv5 real start | |
| layer { | |
| bottom: "pool4" | |
| top: "im2col_conv5_1" | |
| name: "im2col_conv5_1" | |
| type: "Im2col" | |
| convolution_param { | |
| num_output: 512 kernel_size: 3 pad: 2 dilation: 2 | |
| } | |
| } | |
| layer { | |
| bottom: "im2col_conv5_1" | |
| bottom: "exp_emb_dist5_1_512chan" | |
| top: "im2col_conv5_1_hit" | |
| name: "im2col_conv5_1_hit" | |
| type: "Eltwise" | |
| eltwise_param { operation: PROD } | |
| } | |
| layer { | |
| bottom: "im2col_conv5_1_hit" | |
| top: "conv5_1" | |
| name: "conv5_1" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 512 kernel_size: 3 pad: 2 dilation: 2 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "conv5_1" | |
| top: "conv5_1" | |
| name: "relu5_1" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "conv5_1" | |
| top: "im2col_conv5_2" | |
| name: "im2col_conv5_2" | |
| type: "Im2col" | |
| convolution_param { | |
| num_output: 512 kernel_size: 3 pad: 2 dilation: 2 | |
| } | |
| } | |
| layer { | |
| bottom: "im2col_conv5_2" | |
| bottom: "exp_emb_dist5_2_512chan" | |
| top: "im2col_conv5_2_hit" | |
| name: "im2col_conv5_2_hit" | |
| type: "Eltwise" | |
| eltwise_param { operation: PROD } | |
| } | |
| layer { | |
| bottom: "im2col_conv5_2_hit" | |
| top: "conv5_2" | |
| name: "conv5_2" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 512 kernel_size: 3 pad: 2 dilation: 2 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "conv5_2" | |
| top: "conv5_2" | |
| name: "relu5_2" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "conv5_2" | |
| top: "im2col_conv5_3" | |
| name: "im2col_conv5_3" | |
| type: "Im2col" | |
| convolution_param { | |
| num_output: 512 kernel_size: 3 pad: 2 dilation: 2 | |
| } | |
| } | |
| layer { | |
| bottom: "im2col_conv5_3" | |
| bottom: "exp_emb_dist5_3_512chan" | |
| top: "im2col_conv5_3_hit" | |
| name: "im2col_conv5_3_hit" | |
| type: "Eltwise" | |
| eltwise_param { operation: PROD } | |
| } | |
| layer { | |
| bottom: "im2col_conv5_3_hit" | |
| top: "conv5_3" | |
| name: "conv5_3" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 512 kernel_size: 3 pad: 2 dilation: 2 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "conv5_3" | |
| top: "conv5_3" | |
| name: "relu5_3" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "conv5_3" | |
| top: "pool5" | |
| name: "pool5" | |
| type: "Pooling" | |
| pooling_param { | |
| pool: MAX | |
| kernel_size: 3 | |
| stride: 1 | |
| pad: 1 | |
| } | |
| } | |
| layer { bottom: "pool5" top: "pool5a" name: "pool5a" type: "Pooling" | |
| pooling_param { | |
| pool: AVE | |
| kernel_size: 3 pad: 1 | |
| stride: 1 | |
| } | |
| } | |
| ###### Mask prep for fc6 | |
| layer { | |
| bottom: "emb_shrink8" | |
| top: "emb_dist6" | |
| name: "emb_dist6" | |
| type: "Im2dist" | |
| convolution_param { | |
| kernel_size: 3 pad: 12 dilation: 12 | |
| stride: 1 | |
| } | |
| im2dist_param { norm: L1 } | |
| } | |
| layer { | |
| bottom: "emb_dist6" | |
| top: "weighted_emb_dist6" | |
| name: "weighted_emb_dist6" | |
| type: "Scale" | |
| param { lr_mult: 1 decay_mult: 1 } | |
| scale_param { num_axes: 0 filler { type: "constant" value: 0.05 } bias_term: false } | |
| } | |
| layer { bottom: "weighted_emb_dist6" top: "abs_emb_dist6" name: "abs_emb_dist6" type: "AbsVal"} | |
| layer { bottom: "abs_emb_dist6" top: "exp_emb_dist6" name: "exp_emb_dist6" type: "Exp" exp_param { scale: -1 }} | |
| layer { bottom: "exp_emb_dist6" top: "exp_emb_dist6_512chan" name: "exp_emb_dist6_512chan" type: "Tile" tile_param { axis: 1 tiles: 512 }} | |
| ## fc6 real start | |
| layer { | |
| bottom: "pool5a" | |
| top: "im2col_fc6" | |
| name: "im2col_fc6" | |
| type: "Im2col" | |
| convolution_param { | |
| num_output: 1024 kernel_size: 3 pad: 12 dilation: 12 | |
| } | |
| } | |
| layer { | |
| bottom: "im2col_fc6" | |
| bottom: "exp_emb_dist6_512chan" | |
| top: "im2col_fc6_hit" | |
| name: "im2col_fc6_hit" | |
| type: "Eltwise" | |
| eltwise_param { operation: PROD } | |
| } | |
| layer { | |
| bottom: "im2col_fc6_hit" | |
| top: "fc6" | |
| name: "fc6" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 1024 kernel_size: 3 pad: 12 dilation: 12 | |
| bottom_is_im2col: true | |
| } | |
| } | |
| layer { | |
| bottom: "fc6" | |
| top: "fc6" | |
| name: "relu6" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "fc6" | |
| top: "fc6" | |
| name: "drop6" | |
| type: "Dropout" | |
| dropout_param { | |
| dropout_ratio: 0.5 | |
| } | |
| } | |
| layer { | |
| bottom: "fc6" | |
| top: "fc7" | |
| name: "fc7" | |
| type: "Convolution" | |
| param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 1024 | |
| kernel_size: 1 | |
| } | |
| } | |
| layer { | |
| bottom: "fc7" | |
| top: "fc7" | |
| name: "relu7" | |
| type: "ReLU" | |
| } | |
| layer { | |
| bottom: "fc7" | |
| top: "fc7" | |
| name: "drop7" | |
| type: "Dropout" | |
| dropout_param { | |
| dropout_ratio: 0.5 | |
| } | |
| } | |
| layer { | |
| bottom: "fc7" | |
| top: "fc8_voc12" | |
| name: "fc8_voc12" | |
| type: "Convolution" | |
| param { lr_mult: 10 decay_mult: 1 } param { lr_mult: 20 decay_mult: 0 } | |
| convolution_param { | |
| num_output: 21 | |
| kernel_size: 1 | |
| weight_filler { | |
| type: "gaussian" | |
| std: 0.01 | |
| } | |
| bias_filler { | |
| type: "constant" | |
| value: 0 | |
| } | |
| } | |
| } | |
| layer { | |
| bottom: "label" | |
| top: "label_shrink" | |
| name: "label_shrink" | |
| type: "Interp" | |
| interp_param { | |
| shrink_factor: 8 | |
| pad_beg: 0 | |
| pad_end: 0 | |
| } | |
| } | |
| layer { | |
| name: "loss" | |
| type: "SoftmaxWithLoss" | |
| bottom: "fc8_voc12" | |
| bottom: "label_shrink" | |
| loss_param { | |
| ignore_label: 255 | |
| } | |
| } |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment