开源软件名称（OpenSource Name）：UjjwalSaxena/Automold--Road-Augmentation-Library开源软件地址(OpenSource Url)：https://github.com/UjjwalSaxena/Automold--Road-Augmentation-Library开源编程语言(OpenSource Language)：Jupyter Notebook 99.9%开源软件介绍(OpenSource Introduction)：AutomoldThere are various types of image augmentations done to increase the image corpus for training neural networks. However for training CNNs to drive some special road conditions are required. These can be random gravels of the road or maybe snow. Rain and fog also reduce the visibility to a great extent. Automold helps in addressing these challenges and augments road images to have various weather and road conditions. Importing road augmentation library Automold and helper functions libraryimport Automold as am
import Helpers as hpLet's load up some images firstpath='./test_augmentation/*.jpg'
images= hp.load_images(path)visualize function helps in displaying images easily without requiring you to write the whole code.hp.visualize(images, column=3, fig_size=(20,10))
Checking out the brighten functionbrightenparameters image: image or image list brightness_coeff amount of brightness (0<=brightness_coeff<=1), default: random bright_images= am.brighten(images[0:3]) ## if brightness_coeff is undefined brightness is random in each image
hp.visualize(bright_images, column=3)
bright_images= am.brighten(images[0:3], brightness_coeff=0.7) ## brightness_coeff is between 0.0 and 1.0
hp.visualize(bright_images, column=3)
Let's darken a few images nowdarkenparameters image: image or image list darkness_coeff amount of darkness (0<=darkness_coeff<=1), default: random dark_images= am.darken(images[0:3]) ## if darkness_coeff is undefined darkness is random in each image
hp.visualize(dark_images, column=3)
dark_images= am.darken(images[0:3], darkness_coeff=0.7) ## darkness_coeff is between 0.0 and 1.0
hp.visualize(dark_images, column=3)
But what if you just want some random brightness or darkness in the images. Well try out the random_brightness function which receives an image or an image arrayrandom_brightnessparameters image: image or image list dark_bright_images= am.random_brightness(images[4:7])
hp.visualize(dark_bright_images, column=3)
What about adding some shadows to the images.add_shadowParameters image: image or image list no_of_shadows: no. of shadows, default: 1 rectangular_roi: (top-left x, top-left y, bottom-right x, bottom right y), default: lower half of image shadow_dimension: no. of sides of the shadows (3<=shadow_dimension<=10), default: random shadowy_images= am.add_shadow(images[4:7])
hp.visualize(shadowy_images, column=3)
shadowy_images= am.add_shadow(images[4:7], no_of_shadows=2, shadow_dimension=8)
hp.visualize(shadowy_images, column=3)
Now let's add some snowadd_snowparameters image: image or image list snow_coeff: amount of snow (0<=snow_coeff<=1), default: random snowy_images= am.add_snow(images[4:7]) ##randomly add snow
hp.visualize(snowy_images, column=3)
snowy_images= am.add_snow(images[4:7], snow_coeff=0.3)
hp.visualize(snowy_images, column=3)
snowy_images= am.add_snow(images[4:7], snow_coeff=0.8)
hp.visualize(snowy_images, column=3)
and now some rainadd_rainparameters image: image or image list slant: deviation of rain from normal (-20<=slant<=20), default: random drop_length: length of the drop (0<=drop_length<=100), default: 20 (pixels) drop_width: width of the drop (1<=drop_width<=5), default: 1 drop_color: color of droplets, default: (200,200,200) rain_type: values in 'drizzle','heavy','torrential', default: 'None' rainy_images= am.add_rain(images[4:7])
hp.visualize(rainy_images, column=3)
rainy_images= am.add_rain(images[4:7], rain_type='heavy', slant=20)
hp.visualize(rainy_images, column=3)
rainy_images= am.add_rain(images[4:7], rain_type='torrential')
hp.visualize(rainy_images, column=3)
Note: drop_length and drop_width values are overriden when rain_type is not None add_fogparameters image: image or image list fog_coeff: amount of fog (0<=fog_coeff<=1), default: random foggy_images= am.add_fog(images[4:7])
hp.visualize(foggy_images, column=3)
foggy_images= am.add_fog(images[4:7], fog_coeff=0.4)
hp.visualize(foggy_images, column=3)
foggy_images= am.add_fog(images[4:7], fog_coeff=0.9)
hp.visualize(foggy_images, column=3)
what about some gravels on the road now ?add_gravelparameters image: image or image list rectangular_roi: (top-left x, top-left y, bottom-right x, bottom right y), default: lower 3/4th of image no_of_patches: no. of gravel patches required, default: 8 bad_road_images= am.add_gravel(images[4:7])
hp.visualize(bad_road_images, column=3)
bad_road_images= am.add_gravel(images[4:7], rectangular_roi=(700,550,1280,720),no_of_patches=20) ##too much gravels on right
hp.visualize(bad_road_images, column=3)
add_sun_flareparameters image: image or image list flare_center: center coordinates (x,y) of the source, default: random angle: angle of flare in radians, default: random no_of_flare_circles: no. of secondary flare circles (0<=no_of_flare_circles<=20), default: 8 src_radius: radius of the primary flare source, default: 400 (pixels) src_color: rgb color of the flare source and secondary circles, default: (255,255,255)) flare_images= am.add_sun_flare(images[4:7])
hp.visualize(flare_images, column=3)
import math
flare_images= am.add_sun_flare(images[4:7], flare_center=(100,100), angle=-math.pi/4) ## fixed src center
hp.visualize(flare_images, column=3)
add_speedparameters image: image or image list speed_coeff: amount of speed (0<=speed_coeff<=1), default: random speedy_images= am.add_speed(images[1:4]) ##random speed
hp.visualize(speedy_images, column=3)
speedy_images= am.add_speed(images[1:4], speed_coeff=0.9) ##random speed
hp.visualize(speedy_images, column=3)
add_autumnparameters image: image or image list fall_images= am.add_autumn(images[0:3])
hp.visualize(fall_images, column=3)
fliphparameters image: image or image list flipped_images= am.fliph(images[0:3])
hp.visualize(flipped_images, column=3)
flipvparameters image: image or image list flipped_images= am.flipv(images[0:3])
hp.visualize(flipped_images, column=3)
random_flipparameters image: image or image list flipped_images= am.random_flip(images[0:3])
hp.visualize(flipped_images, column=3)
add_manholeparameters image: image or image list center: center of the ellipse (x,y), default: bottom center of the image color: rgb tuple, default: if type parameter not defined: (67,70,75), else: default color mentioned in type. height: vertical dimension of the hole, int , default: 25th portion of the image height. width: horizontal dimension of the hole, int, default: 3/25th portion of the image height. type: values in 'closed','open', default: 'closed' manhole_images= am.add_manhole(images[0:3])
hp.visualize(manhole_images, column=3)
correct_exposureparameters image: image or image list exposure_images= am.correct_exposure(images[0:3])
hp.visualize(exposure_images, column=3)If a series of augmentations is required from above types augment_random function can be handyaugment_randomimage: image or image list aug_types: list of Automold functions, eg: ['add_snow','add_rain'], default: all aug functions are executed volume: 'same' or 'expand', default: expand aug_images= am.augment_random(images[4:6], volume='same') ##2 random augmentations applied on both images
hp.visualize(aug_images,column=3,fig_size=(20,20))
aug_images= am.augment_random(images[4:6], volume='expand') ##all aug_types are applied in both images
hp.visualize(aug_images,column=3,fig_size=(20,20))
aug_images= am.augment_random(images[4:6], aug_types=['add_sun_flare','add_speed','add_autumn'], volume='expand') ##all aug_types are applied in both images
hp.visualize(aug_images,column=3,fig_size=(20,10))
aug_images= am.augment_random(images[4:6], aug_types=['add_sun_flare','add_speed','add_autumn'], volume='same') ##2 random aug_types are applied in both images
hp.visualize(aug_images,column=3,fig_size=(20,10))
Performance statistics全部评论
专题导读
上一篇：ThomasKing2014/ELF-ARM-HOOK-Library: It's very similiar to Substrate. But I ...发布时间：2022-08-15下一篇：PHPMailer/PHPMailer: The classic email sending library for PHP发布时间：2022-08-15热门推荐
热门话题
 阅读排行榜
|
请发表评论