Caffe Augmentation Extension
This is a modified caffe fork (version of 2017/3/10) with ImageData layer data augmentation, which is based on:
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
}
}
扫一扫
1891
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
