guoqiang0148666 / caffe-augmentation

Data Augmentation for Caffe

Caffe Augmentation Extension

This is a modified caffe fork (version of 2017/3/10) with ImageData layer data augmentation, which is based on:

@kevinlin311tw's caffe-augmentation, @ChenlongChen's caffe-windows, @ShaharKatz's Caffe-Data-Augmentation, @senecaur's caffe-rta. @kostyaev's caffe-augmentation

min_side_min nad min_side_max are added for random cropping while keeping the aspect ratio, as mentioned in "Deep Residual Learning for Image Recognition"(http://arxiv.org/abs/1512.03385)

and all functions

• min_side - resize and crop preserving aspect ratio, default 0 (disabled);
• max_rotation_angle - max angle for an image rotation, default 0;
• contrast_brightness_adjustment - enable/disable contrast adjustment, default false;
• smooth_filtering - enable/disable smooth filterion, default false;
• min_contrast - min contrast multiplier (min alpha), default 0.8;
• max_contrast - min contrast multiplier (max alpha), default 1.2;
• max_brightness_shift - max brightness shift in positive and negative directions (beta), default 5;
• max_smooth - max smooth multiplier, default 6;
• max_color_shift - max color shift along RGB axes
• apply_probability - how often every transformation should be applied, default 0.5;
• debug_params - enable/disable printing tranformation parameters, default false;

from @kostyaev's caffe-augmentation are kept with slightly modifications:

How to use

You could specify your network prototxt as:

layer {
name: "data"
type: "ImageData"
top: "data"
top: "label"
include {
  phase: TRAIN
}
transform_param {
    mirror: true
    contrast_brightness_adjustment: true
    smooth_filtering: true
    min_side_min: 256
    min_side_max: 480
    crop_size: 224
    mean_file: "imagenet_mean.binaryproto"
    min_contrast: 0.8
    max_contrast: 1.2
    max_smooth: 6
    apply_probability: 0.5
    max_color_shift: 20
    debug_params: false
}
image_data_param {
  source: "train_list.txt"
  batch_size: 64
}
}

while in testing phase:

layer {
name: "data"
type: "ImageData"
top: "data"
top: "label"
include {
  phase: TEST
}
transform_param {
    mirror: false
    min_side: 256
    crop_size: 224
    mean_file: "imagenet_mean.binaryproto"
}
image_data_param {
  source: "test_list.txt"
  batch_size: 32
}
}