zhaoweicai
commented
7 years ago This layer pre-processes the input images, and outputs data and label blobs for later layers. The output image data is at a single scale, and the output label data is at multiple scales for multi-scale proposal generation training.
@zhaoweicai What is the function of the image_gt_data_layer ？And what does the outputs in this layer mean ？
layer { top: "data" top: "label_1_5x5" top: "label_1_7x7" top: "label_2_5x5" top: "label_2_7x7" top: "label_3_5x5" top: "label_3_7x7" top: "label_4_5x5" name: "data" type: "ImageGtData" include { phase: TRAIN } transform_param { mirror: true mean_value: 104 mean_value: 117 mean_value: 123 } image_gt_data_param { source: "../../../data/kitti/window_files/mscnn_window_file_kitti_vehicle_train.txt" batch_size: 4 coord_num: 4 resize_width: 1280 resize_height: 384 crop_width: 512 crop_height: 384 min_gt_height: 25 downsample_rate: 8 downsample_rate: 8 downsample_rate: 16 downsample_rate: 16 downsample_rate: 32 downsample_rate: 32 downsample_rate: 64 field_w: 40 field_w: 56 field_w: 80 field_w: 112 field_w: 160 field_w: 224 field_w: 320 field_h: 40 field_h: 56 field_h: 80 field_h: 112 field_h: 160 field_h: 224 field_h: 320 fg_threshold: 0.5 do_multiple_scale: true min_scale: 40 max_scale: 320 shuffle: true } }