jackweiwang
commented
6 years ago I solved it， change mean_value 1 channel to 3 channel.
Closed jackweiwang closed 6 years ago
jackweiwang
commented
6 years ago I solved it， change mean_value 1 channel to 3 channel.
WillDamon
commented
6 years ago @jackweiwang Hello，I meet the same problem as you showed when I want to train on UCF101 dataset, but I do not know what do you mean by "changing mean_value 1 channel to 3 channel"? Can you explain in detail for me? Thank you so much!
jackweiwang
commented
6 years ago @jackweiwang Hello，I meet the same problem as you showed when I want to train on UCF101 dataset, but I do not know what do you mean by "changing mean_value 1 channel to 3 channel"? Can you explain in detail for me? Thank you so much!
The author use 16 frames of mean_value for training, so i have modified this file according to my own sample dataset. changed mean_value: [104] mean_value: [117] mean_value: [123] to mean_value: [104,117,123], you need to look at the code here and change sample dataset. I hope I can help you.
WillDamon
commented
6 years ago @jackweiwang Hello, I retried as what you told, But it is still not OK, I am so confused...... Maybe you can see my ECO_Lite.prototxt, it is just as follows:
name: "o3d" mem_param { optimize_test: true optimize_train: true }
layer { name: "data" type: "VideoData" top: "data" top: "label" video_data_param { source: "/SSD2TB/wangwei/ECO-efficient-video-understanding/ucf101_train.txt" batch_size: 17 new_length: 1 num_segments: 16 modality: RGB shuffle: true name_pattern: "%04d.jpg" } transform_param{ crop_size: 224 mirror: true fix_crop: true more_fix_crop: true multi_scale: true max_distort: 1 scale_ratios:[1,.875,.75, .66] is_flow: false mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] } include: { phase: TRAIN } } layer { name: "data" type: "VideoData" top: "data" top: "label" video_data_param { source: "/SSD2TB/wangwei/ECO-efficient-video-understanding/ucf101_test.txt" batch_size: 1 new_length: 1 num_segments: 16 modality: RGB name_pattern: "%04d.jpg" } transform_param{ crop_size: 224 mirror: false mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123] mean_value: [104,117,123]
} include: { phase: TEST } } layer { name: "reshape_data" type: "Reshape" bottom: "data" top: "reshape_data" reshape_param { shape { dim: -1 dim: 3 dim: 224 dim: 224 } } }
layer { name: "conv1_7x7_s2" type: "Convolution" bottom: "reshape_data" top: "conv1_7x7_s2" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 64 pad: 3 kernel_size: 7 stride: 2 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "conv1_7x7_s2_bn" type: "BN" bottom: "conv1_7x7_s2" top: "conv1_7x7_s2_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "conv1_relu_7x7_inp" type: "ReLU" bottom: "conv1_7x7_s2_bn" top: "conv1_7x7_s2_bn" } layer { name: "pool1_3x3_s2" type: "Pooling" bottom: "conv1_7x7_s2_bn" top: "pool1_3x3_s2" pooling_param { pool: MAX kernel_size: 3 stride: 2 } } layer { name: "conv2_3x3_reduce" type: "Convolution" bottom: "pool1_3x3_s2" top: "conv2_3x3_reduce" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 64 kernel_size: 1 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "conv2_3x3_reduce_bn" type: "BN" bottom: "conv2_3x3_reduce" top: "conv2_3x3_reduce_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "conv2_relu_3x3_reduce_inp" type: "ReLU" bottom: "conv2_3x3_reduce_bn" top: "conv2_3x3_reduce_bn" } layer { name: "conv2_3x3" type: "Convolution" bottom: "conv2_3x3_reduce_bn" top: "conv2_3x3" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 192 pad: 1 kernel_size: 3 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "conv2_3x3_bn" type: "BN" bottom: "conv2_3x3" top: "conv2_3x3_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "conv2_relu_3x3_inp" type: "ReLU" bottom: "conv2_3x3_bn" top: "conv2_3x3_bn" } layer { name: "pool2_3x3_s2" type: "Pooling" bottom: "conv2_3x3_bn" top: "pool2_3x3_s2" pooling_param { pool: MAX kernel_size: 3 stride: 2 } } layer { name: "inception_3a_1x1" type: "Convolution" bottom: "pool2_3x3_s2" top: "inception_3a_1x1" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 64 kernel_size: 1 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3a_1x1_bn" type: "BN" bottom: "inception_3a_1x1" top: "inception_3a_1x1_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3a_relu_1x1_inp" type: "ReLU" bottom: "inception_3a_1x1_bn" top: "inception_3a_1x1_bn" } layer { name: "inception_3a_3x3_reduce" type: "Convolution" bottom: "pool2_3x3_s2" top: "inception_3a_3x3_reduce" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 64 kernel_size: 1 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3a_3x3_reduce_bn" type: "BN" bottom: "inception_3a_3x3_reduce" top: "inception_3a_3x3_reduce_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3a_relu_3x3_reduce_inp" type: "ReLU" bottom: "inception_3a_3x3_reduce_bn" top: "inception_3a_3x3_reduce_bn" } layer { name: "inception_3a_3x3" type: "Convolution" bottom: "inception_3a_3x3_reduce_bn" top: "inception_3a_3x3" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 64 pad: 1 kernel_size: 3 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3a_3x3_bn" type: "BN" bottom: "inception_3a_3x3" top: "inception_3a_3x3_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3a_relu_3x3_inp" type: "ReLU" bottom: "inception_3a_3x3_bn" top: "inception_3a_3x3_bn" } layer { name: "inception_3a_double_3x3_reduce" type: "Convolution" bottom: "pool2_3x3_s2" top: "inception_3a_double_3x3_reduce" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 64 kernel_size: 1 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3a_double_3x3_reduce_bn" type: "BN" bottom: "inception_3a_double_3x3_reduce" top: "inception_3a_double_3x3_reduce_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3a_relu_double_3x3_reduce_inp" type: "ReLU" bottom: "inception_3a_double_3x3_reduce_bn" top: "inception_3a_double_3x3_reduce_bn" } layer { name: "inception_3a_double_3x3_1" type: "Convolution" bottom: "inception_3a_double_3x3_reduce_bn" top: "inception_3a_double_3x3_1" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 96 pad: 1 kernel_size: 3 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3a_double_3x3_1_bn" type: "BN" bottom: "inception_3a_double_3x3_1" top: "inception_3a_double_3x3_1_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3a_relu_double_3x3_1_inp" type: "ReLU" bottom: "inception_3a_double_3x3_1_bn" top: "inception_3a_double_3x3_1_bn" } layer { name: "inception_3a_double_3x3_2" type: "Convolution" bottom: "inception_3a_double_3x3_1_bn" top: "inception_3a_double_3x3_2" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 96 pad: 1 kernel_size: 3 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3a_double_3x3_2_bn" type: "BN" bottom: "inception_3a_double_3x3_2" top: "inception_3a_double_3x3_2_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3a_relu_double_3x3_2_inp" type: "ReLU" bottom: "inception_3a_double_3x3_2_bn" top: "inception_3a_double_3x3_2_bn" } layer { name: "inception_3a_pool" type: "Pooling" bottom: "pool2_3x3_s2" top: "inception_3a_pool" pooling_param { pool: AVE kernel_size: 3 stride: 1 pad: 1 } } layer { name: "inception_3a_pool_proj" type: "Convolution" bottom: "inception_3a_pool" top: "inception_3a_pool_proj" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 32 kernel_size: 1 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3a_pool_proj_bn" type: "BN" bottom: "inception_3a_pool_proj" top: "inception_3a_pool_proj_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3a_relu_pool_proj_inp" type: "ReLU" bottom: "inception_3a_pool_proj_bn" top: "inception_3a_pool_proj_bn" } layer { name: "inception_3a_output" type: "Concat" bottom: "inception_3a_1x1_bn" bottom: "inception_3a_3x3_bn" bottom: "inception_3a_double_3x3_2_bn" bottom: "inception_3a_pool_proj_bn" top: "inception_3a_output" } layer { name: "inception_3b_1x1" type: "Convolution" bottom: "inception_3a_output" top: "inception_3b_1x1" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 64 kernel_size: 1 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3b_1x1_bn" type: "BN" bottom: "inception_3b_1x1" top: "inception_3b_1x1_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3b_relu_1x1_inp" type: "ReLU" bottom: "inception_3b_1x1_bn" top: "inception_3b_1x1_bn" } layer { name: "inception_3b_3x3_reduce" type: "Convolution" bottom: "inception_3a_output" top: "inception_3b_3x3_reduce" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 64 kernel_size: 1 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3b_3x3_reduce_bn" type: "BN" bottom: "inception_3b_3x3_reduce" top: "inception_3b_3x3_reduce_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3b_relu_3x3_reduce_inp" type: "ReLU" bottom: "inception_3b_3x3_reduce_bn" top: "inception_3b_3x3_reduce_bn" } layer { name: "inception_3b_3x3" type: "Convolution" bottom: "inception_3b_3x3_reduce_bn" top: "inception_3b_3x3" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 96 pad: 1 kernel_size: 3 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3b_3x3_bn" type: "BN" bottom: "inception_3b_3x3" top: "inception_3b_3x3_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3b_relu_3x3_inp" type: "ReLU" bottom: "inception_3b_3x3_bn" top: "inception_3b_3x3_bn" } layer { name: "inception_3b_double_3x3_reduce" type: "Convolution" bottom: "inception_3a_output" top: "inception_3b_double_3x3_reduce" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 64 kernel_size: 1 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3b_double_3x3_reduce_bn" type: "BN" bottom: "inception_3b_double_3x3_reduce" top: "inception_3b_double_3x3_reduce_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3b_relu_double_3x3_reduce_inp" type: "ReLU" bottom: "inception_3b_double_3x3_reduce_bn" top: "inception_3b_double_3x3_reduce_bn" } layer { name: "inception_3b_double_3x3_1" type: "Convolution" bottom: "inception_3b_double_3x3_reduce_bn" top: "inception_3b_double_3x3_1" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 96 pad: 1 kernel_size: 3 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3b_double_3x3_1_bn" type: "BN" bottom: "inception_3b_double_3x3_1" top: "inception_3b_double_3x3_1_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3b_relu_double_3x3_1_inp" type: "ReLU" bottom: "inception_3b_double_3x3_1_bn" top: "inception_3b_double_3x3_1_bn" } layer { name: "inception_3b_double_3x3_2" type: "Convolution" bottom: "inception_3b_double_3x3_1_bn" top: "inception_3b_double_3x3_2" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 96 pad: 1 kernel_size: 3 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3b_double_3x3_2_bn" type: "BN" bottom: "inception_3b_double_3x3_2" top: "inception_3b_double_3x3_2_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3b_relu_double_3x3_2_inp" type: "ReLU" bottom: "inception_3b_double_3x3_2_bn" top: "inception_3b_double_3x3_2_bn" } layer { name: "inception_3b_pool" type: "Pooling" bottom: "inception_3a_output" top: "inception_3b_pool" pooling_param { pool: AVE kernel_size: 3 stride: 1 pad: 1 } } layer { name: "inception_3b_pool_proj" type: "Convolution" bottom: "inception_3b_pool" top: "inception_3b_pool_proj" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 64 kernel_size: 1 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3b_pool_proj_bn" type: "BN" bottom: "inception_3b_pool_proj" top: "inception_3b_pool_proj_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3b_relu_pool_proj_inp" type: "ReLU" bottom: "inception_3b_pool_proj_bn" top: "inception_3b_pool_proj_bn" } layer { name: "inception_3b_output" type: "Concat" bottom: "inception_3b_1x1_bn" bottom: "inception_3b_3x3_bn" bottom: "inception_3b_double_3x3_2_bn" bottom: "inception_3b_pool_proj_bn" top: "inception_3b_output" }
layer { name: "inception_3c_double_3x3_reduce" type: "Convolution" bottom: "inception_3b_output" top: "inception_3c_double_3x3_reduce" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 64 kernel_size: 1 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3c_double_3x3_reduce_bn" type: "BN" bottom: "inception_3c_double_3x3_reduce" top: "inception_3c_double_3x3_reduce_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3c_relu_double_3x3_reduce_inp" type: "ReLU" bottom: "inception_3c_double_3x3_reduce_bn" top: "inception_3c_double_3x3_reduce_bn" } layer { name: "inception_3c_double_3x3_1" type: "Convolution" bottom: "inception_3c_double_3x3_reduce_bn" top: "inception_3c_double_3x3_1" param { lr_mult: 1.0 decay_mult: 1.0 } param { lr_mult: 1.0 decay_mult: 2.0 } convolution_param { num_output: 96 pad: 1 kernel_size: 3 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3c_double_3x3_1_bn" type: "BN" bottom: "inception_3c_double_3x3_1" top: "inception_3c_double_3x3_1_bn" param { lr_mult: 1.0 decay_mult: 0.0 } param { lr_mult: 1.0 decay_mult: 0.0 } bn_param { slope_filler { type: "constant" value: 1.0 } bias_filler { type: "constant" value: 0.0 } } } layer { name: "inception_3c_relu_double_3x3_1_inp" type: "ReLU" bottom: "inception_3c_double_3x3_1_bn" top: "inception_3c_double_3x3_1_bn" }
layer { name: "r2Dto3D" type: "Reshape" bottom: "inception_3c_double_3x3_1_bn" top: "res2b_bn_pre" reshape_param{ shape { dim: -1 dim: 16 dim: 96 dim: 28 dim: 28 } } } layer { name: "Transpose1" type: "Permute" bottom: "res2b_bn_pre" top: "res2b_bn" permute_param { order: [0,2,1,3,4] } }
########################################### Res3a ##########################################
layer { name: "res3a_2n" bottom: "res2b_bn" top: "res3a" type: "Convolution" convolution_param { num_output: 128 pad: [1, 1, 1] kernel_size: [3, 3, 3] stride: [1, 1, 1] weight_filler{ type: "xavier" } bias_filler{ type: "constant" value: 0 } } param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } }
layer { name: "res3a_bn" bottom: "res3a" top: "res3a_bn" type: "BN" param { lr_mult: 1 decay_mult: 0 } param { lr_mult: 1 decay_mult: 0 } bn_param { frozen: false slope_filler { type: "constant" value: 1 } bias_filler { type: "constant" value: 0 } } } layer { name: "res3a_relu" bottom: "res3a_bn" top: "res3a_bn" type: "ReLU" } ############################################ Res3b #################################### layer { name: "res3b_1" bottom: "res3a_bn" top: "res3b_1" type: "Convolution" convolution_param { num_output: 128 pad: [1, 1, 1] kernel_size: [3, 3, 3] stride: [1, 1, 1] weight_filler{ type: "xavier" } bias_filler{ type: "constant" value: 0 } } param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } } layer { name: "res3b_1_bn" bottom: "res3b_1" top: "res3b_1_bn" type: "BN" param { lr_mult: 1 decay_mult: 0 } param { lr_mult: 1 decay_mult: 0 } bn_param { frozen: false slope_filler { type: "constant" value: 1 } bias_filler { type: "constant" value: 0 } } } layer { name: "res3b_1_relu" bottom: "res3b_1_bn" top: "res3b_1_bn" type: "ReLU" } layer { name: "res3b_2" bottom: "res3b_1_bn" top: "res3b_2" type: "Convolution" convolution_param { num_output: 128 pad: [1, 1, 1] kernel_size: [3, 3, 3] stride: [1, 1, 1] weight_filler{ type: "xavier" } bias_filler{ type: "constant" value: 0 } } param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } } layer { name: "res3b" bottom: "res3b_2" bottom: "res3a" top: "res3b" type: "Eltwise" } layer { name: "res3b_bn" bottom: "res3b" top: "res3b_bn" type: "BN" param { lr_mult: 1 decay_mult: 0 } param { lr_mult: 1 decay_mult: 0 } bn_param { frozen: false slope_filler { type: "constant" value: 1 } bias_filler { type: "constant" value: 0 } } } layer { name: "res3b_relu" bottom: "res3b_bn" top: "res3b_bn" type: "ReLU" }
######################################### res4a ################################################# layer { name: "res4a_1" bottom: "res3b_bn" top: "res4a_1" type: "Convolution" convolution_param { num_output: 256 pad: [1, 1, 1] kernel_size: [3, 3, 3] stride: [2, 2, 2] weight_filler{ type: "xavier" } bias_filler{ type: "constant" value: 0 } } param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } } layer { name: "res4a_1_bn" bottom: "res4a_1" top: "res4a_1_bn" type: "BN" param { lr_mult: 1 decay_mult: 0 } param { lr_mult: 1 decay_mult: 0 } bn_param { frozen: false slope_filler { type: "constant" value: 1 } bias_filler { type: "constant" value: 0 } } } layer { name: "res4a_1_relu" bottom: "res4a_1_bn" top: "res4a_1_bn" type: "ReLU" } layer { name: "res4a_2" bottom: "res4a_1_bn" top: "res4a_2" type: "Convolution" convolution_param { num_output: 256 pad: [1, 1, 1] kernel_size: [3, 3, 3] stride: [1, 1, 1] weight_filler{ type: "xavier" } bias_filler{ type: "constant" value: 0 } } param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } } layer { name: "res4a_down" bottom: "res3b_bn" top: "res4a_down" type: "Convolution" convolution_param { num_output: 256 pad: [1, 1, 1] kernel_size: [3, 3, 3] stride: [2, 2, 2] weight_filler{ type: "xavier" } bias_filler{ type: "constant" value: 0 } } param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } } layer { name: "res4a" bottom: "res4a_2" bottom: "res4a_down" top: "res4a" type: "Eltwise" } layer { name: "res4a_bn" bottom: "res4a" top: "res4a_bn" type: "BN" param { lr_mult: 1 decay_mult: 0 } param { lr_mult: 1 decay_mult: 0 } bn_param { frozen: false slope_filler { type: "constant" value: 1 } bias_filler { type: "constant" value: 0 } } } layer { name: "res4a_relu" bottom: "res4a_bn" top: "res4a_bn" type: "ReLU" } ################################################ Res4b ########################################################### layer { name: "res4b_1" bottom: "res4a_bn" top: "res4b_1" type: "Convolution" convolution_param { num_output: 256 pad: [1, 1, 1] kernel_size: [3, 3, 3] stride: [1, 1, 1] weight_filler{ type: "xavier" } bias_filler{ type: "constant" value: 0 } } param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } } layer { name: "res4b_1_bn" bottom: "res4b_1" top: "res4b_1_bn" type: "BN" param { lr_mult: 1 decay_mult: 0 } param { lr_mult: 1 decay_mult: 0 } bn_param { frozen: false slope_filler { type: "constant" value: 1 } bias_filler { type: "constant" value: 0 } } } layer { name: "res4b_1_relu" bottom: "res4b_1_bn" top: "res4b_1_bn" type: "ReLU" } layer { name: "res4b_2" bottom: "res4b_1_bn" top: "res4b_2" type: "Convolution" convolution_param { num_output: 256 pad: [1, 1, 1] kernel_size: [3, 3, 3] stride: [1, 1, 1] weight_filler{ type: "xavier" } bias_filler{ type: "constant" value: 0 } } param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } } layer { name: "res4b" bottom: "res4b_2" bottom: "res4a" top: "res4b" type: "Eltwise" } layer { name: "res4b_bn" bottom: "res4b" top: "res4b_bn" type: "BN" param { lr_mult: 1 decay_mult: 0 } param { lr_mult: 1 decay_mult: 0 } bn_param { frozen: false slope_filler { type: "constant" value: 1 } bias_filler { type: "constant" value: 0 } } } layer { name: "res4b_relu" top: "res4b_bn" bottom: "res4b_bn" type: "ReLU" } ######################################## Res5a ######################################## layer { name: "res5a_1" bottom: "res4b_bn" top: "res5a_1" type: "Convolution" convolution_param { num_output: 512 pad: [1, 1, 1] kernel_size: [3, 3, 3] stride: [2, 2, 2] weight_filler{ type: "xavier" } bias_filler{ type: "constant" value: 0 } } param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } } layer { name: "res5a_1_bn" bottom: "res5a_1" top: "res5a_1_bn" type: "BN" param { lr_mult: 1 decay_mult: 0 } param { lr_mult: 1 decay_mult: 0 } bn_param { frozen: false slope_filler { type: "constant" value: 1 } bias_filler { type: "constant" value: 0 } } } layer { name: "res5a_1_relu" bottom: "res5a_1_bn" top: "res5a_1_bn" type: "ReLU" } layer { name: "res5a_2" bottom: "res5a_1_bn" top: "res5a_2" type: "Convolution" convolution_param { num_output: 512 pad: [1, 1, 1] kernel_size: [3, 3, 3] stride: [1, 1, 1] weight_filler{ type: "xavier" } bias_filler{ type: "constant" value: 0 } } param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } } layer { name: "res5a_down" bottom: "res4b_bn" top: "res5a_down" type: "Convolution" convolution_param { num_output: 512 pad: [1, 1, 1] kernel_size: [3, 3, 3] stride: [2, 2, 2] weight_filler{ type: "xavier" } bias_filler{ type: "constant" value: 0 } } param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } } layer { name: "res5a" bottom: "res5a_2" bottom: "res5a_down" top: "res5a" type: "Eltwise" } layer { name: "res5a_bn" bottom: "res5a" top: "res5a_bn" type: "BN" param { lr_mult: 1 decay_mult: 0 } param { lr_mult: 1 decay_mult: 0 } bn_param { frozen: false slope_filler { type: "constant" value: 1 } bias_filler { type: "constant" value: 0 } } } layer { name: "res5a_relu" bottom: "res5a_bn" top: "res5a_bn" type: "ReLU" } ####################################### Res5b ######################################### layer { name: "res5b_1" bottom: "res5a_bn" top: "res5b_1" type: "Convolution" convolution_param { num_output: 512 pad: [1, 1, 1] kernel_size: [3, 3, 3] stride: [1, 1, 1] weight_filler{ type: "xavier" } bias_filler{ type: "constant" value: 0 } } param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } } layer { name: "res5b_1_bn" bottom: "res5b_1" top: "res5b_1_bn" type: "BN" param { lr_mult: 1 decay_mult: 0 } param { lr_mult: 1 decay_mult: 0 } bn_param { frozen: false slope_filler { type: "constant" value: 1 } bias_filler { type: "constant" value: 0 } } } layer { name: "res5b_1_relu" bottom: "res5b_1_bn" top: "res5b_1_bn" type: "ReLU" } layer { name: "res5b_2" bottom: "res5b_1_bn" top: "res5b_2" type: "Convolution" convolution_param { num_output: 512 pad: [1, 1, 1] kernel_size: [3, 3, 3] stride: [1, 1, 1] weight_filler{ type: "xavier" } bias_filler{ type: "constant" value: 0 } } param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } } layer { name: "res5b" type: "Eltwise" bottom: "res5b_2" bottom: "res5a" top: "res5b" } layer { name: "res5b_bn" bottom: "res5b" top: "res5b_bn" type: "BN" param { lr_mult: 1 decay_mult: 0 } param { lr_mult: 1 decay_mult: 0 } bn_param { frozen: false slope_filler { type: "constant" value: 1 } bias_filler { type: "constant" value: 0 } } } layer { name: "res5b_relu" bottom: "res5b_bn" top: "res5b_bn" type: "ReLU" } ######################################### Global pool ############################################################ layer { name: "global_pool" bottom: "res5b_bn" top: "global_pool" type: "Pooling" pooling_param { pool: AVE kernel_size: [4, 7, 7] stride: [1, 1, 1] } } layer { name: "global_pool_reshape" type: "Reshape" bottom: "global_pool" top: "global_pool_reshape" reshape_param { shape { dim: -1 dim: 512 } } } layer { name: "dropout" type: "Dropout" bottom: "global_pool_reshape" top: "global_pool_reshape" dropout_param { dropout_ratio: 0.6 } } ########################################## FC ####################################### layer { name: "fc8u" type: "InnerProduct" bottom: "global_pool_reshape" top: "fc8" param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } inner_product_param { num_output: 101 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0 } } } layer { name: "loss" type: "SoftmaxWithLoss" bottom: "fc8" bottom: "label" include { phase: TRAIN } top: "loss" } layer { name: "loss" type: "SoftmaxWithLoss" bottom: "fc8" bottom: "label" top: "loss" include { phase: TEST } } layer { name: "top1" type: "Accuracy" bottom: "fc8" bottom: "label" top: "top1" accuracy_param { top_k : 1} include { phase: TEST } } layer { name: "top5" type: "Accuracy" bottom: "fc8" bottom: "label" top: "top5" accuracy_param { top_k: 5 } include { phase: TEST } }
Hi: @mzolfaghari
I was training ucf101 dataset and set ECO_Lite.prototxt parameters `# ----- video/label input ----- for training layer { name: "data" type: "VideoData" top: "data" top: "label" video_data_param { source: "/home/ww/ECO-efficient-video-understanding/datasets/ucf101/train_rgb_split1.txt" batch_size: 10 new_length: 1 num_segments: 16 modality: RGB shuffle: true namepattern: "image%04d.jpg" } transform_param{ crop_size: 224 mirror: true fix_crop: true more_fix_crop: true multi_scale: true max_distort: 1 scale_ratios:[1,.875,.75, .66] is_flow: false mean_value: [104] mean_value: [117] mean_value: [123]
} include: { phase: TRAIN } } layer { name: "data" type: "VideoData" top: "data" top: "label" video_data_param { source: "/home/ww/ECO-efficient-video-understanding/datasets/ucf101/val_rgb_split1.txt" batch_size: 1 new_length: 1 num_segments: 16 modality: RGB namepattern: "image%04d.jpg" } transform_param{ crop_size: 224 mirror: false mean_value: [104] mean_value: [117] mean_value: [123]
} include: { phase: TEST } }` but,when i start execute script used scripts
sh models_ECO_Lite/ucf101/run.shThe following errors are listed:
`I0904 16:34:31.579064 7738 video_data_layer.cpp:37] Opening file: /home/ww/ECO-efficient-video-understanding/datasets/ucf101/val_rgb_split1.txt
I0904 16:34:31.603790 7738 video_data_layer.cpp:54] A total of 41822 videos. Aborted at 1536050071 (unix time) try "date -d @1536050071" if you are using GNU date PC: @ 0x7f235907b44f caffe::VideoDataLayer<>::DataLayerSetUp() SIGFPE (@0x7f235907b44f) received by PID 7738 (TID 0x7f23597a6740) from PID 1493677135; stack trace: @ 0x7f2357ba74b0 (unknown) @ 0x7f235907b44f caffe::VideoDataLayer<>::DataLayerSetUp() @ 0x7f2359061699 caffe::BasePrefetchingDataLayer<>::LayerSetUp() @ 0x7f23591123c5 caffe::Net<>::Init() @ 0x7f23591149b5 caffe::Net<>::Net() @ 0x7f23590ec1fa caffe::Solver<>::InitTestNets() @ 0x7f23590eca66 caffe::Solver<>::Init() @ 0x7f23590ecc26 caffe::Solver<>::Solver() @ 0x412148 caffe::GetSolver<>() @ 0x409576 train() @ 0x406fb0 main @ 0x7f2357b92830 __libc_start_main @ 0x4074e9 _start @ 0x0 (unknown) Floating point exception (core dumped) ` The 'train_rgb_split1.txt' is loaded correctly while 'test_rgb_split1.txt' is not. Please tell me why.
More detailed logs are there log.txt
Can you give me some advice, thanks!