The hardware and bandwidth for this mirror is donated by METANET, the Webhosting and Full Service-Cloud Provider.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]metanet.ch.

Data augmentation

Intro

The fastai library simplifies training fast and accurate neural nets using modern best practices. See the fastai website to get started. The library is based on research into deep learning best practices undertaken at fast.ai, and includes “out of the box” support for vision, text, tabular, and collab (collaborative filtering) models.

Dataset

Data augmentation plays a huge role while working on Computer Vision task. Because the proper image transformation can drastically improve the generalization while building a deep learning model.

Read image:

img = fastai::Image_create('files/cat.jpeg')

Plot it:

img %>% show() %>% plot()
<center>

<img src="images/cat.png" alt="_" style="width: 350px;"/>

</center>

Flipped

img_res = list(img, img$flip_lr())
titles = c('original', 'flipped')

c(fig, axs) %<-% subplots(1,2)

for (i in 1:2) {
  img_res[[i]] %>% show_image(ax = axs[[i]],
               title=titles[i])
}

img %>% plot(dpi = 250)
<center>

<img src="images/flip.png" alt="_" style="width: 350px;"/>

</center>

Dihedral

c(fig, axs) %<-% subplots(2, 4)

for (i in 1:8) {
  show_image(DihedralItem(p = 1.)(img, split_idx = 0), ctx = axs[[i]])
}

img %>% plot(dpi = 250)
<center>

<img src="images/dihedral.png" alt="_" style="width: 350px;"/>

</center>

CropPad

sz = c(300L, 500L, 700L)
size = paste('Size', sz)

c(fig, axs) %<-% subplots(1, 3, figsize = c(12, 4))

for (i in 1:3) {
  show_image(img$crop_pad(sz[i]), ctx = axs[[i]], title = size[i])
}

img %>% plot(dpi = 250)
<center>

<img src="images/crop.png" alt="_" style="width: 350px;"/>

</center>

PadModes

pad_modes = c('border', 'reflection', 'zeros')

c(fig, axs) %<-% subplots(1, 3, figsize = c(12, 4))

for (i in 1:3) {
  show_image(img$crop_pad(c(600L,700L), pad_mode = pad_modes[i]),
             ctx = axs[[i]], title = pad_modes[i])
}

img %>% plot(dpi = 250)
<center>

<img src="images/modes.png" alt="_" style="width: 350px;"/>

</center>

RandomCrop

c(fig, axs) %<-% subplots(1, 3, figsize = c(12, 4))

ff = RandomCrop(100)

for (i in 1:3) {
  show_image(ff(img), ctx = axs[[i]])
}

img %>% plot(dpi = 250)
<center>

<img src="images/random_crop.png" alt="_" style="width: 350px;"/>

</center>

CenterCrop

c(fig, axs) %<-% subplots(1, 3, figsize = c(12, 4))

ff = RandomCrop(100L)

for (i in 1:3) {
  show_image(ff(img, split_idx = 1L), ctx = axs[[i]])
}

img %>% plot(dpi = 250)
<center>

<img src="images/center_crop.png" alt="_" style="width: 350px;"/>

</center>

Resize method

resize = c('squish', 'pad', 'crop')

c(fig, axs) %<-% subplots(1, 3, figsize = c(12, 4))

for (i in 1:3) {
  rsz = Resize(256, method = resize[i])
  show_image(rsz(img, split_idx = 0L), ctx = axs[[i]], title = resize[i])
}

img %>% plot(dpi = 250)
<center>

<img src="images/resize.png" alt="_" style="width: 350px;"/>

</center>

RandomResizeCrop

c(fig, axs) %<-% subplots(3, 3, figsize = c(9, 9))

ff = RandomResizedCrop(100)

for (i in 1:9) {
  show_image(ff(img), ctx = axs[[i]])
}

img %>% plot(dpi = 250)
<center>

<img src="images/random_resize.png" alt="_" style="width: 350px;"/>

</center>

These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.