List of Keras (images) Preprocessing Layer (KPL)

[innat]

Keras Preprocessing Layer (KPL)

A list to track down the image preprocessing layers, filter out from issues. Please, feel free to share the relevant content for KPL.

• CutMix, MixUp augmentation
  • Classification : PR
  • Detection : PR (WIP)
• Mosaic Augmentation - Object Detection
  • Classification
  • Detection
• RandomFlip
  • Classification
  • Detection
• CutOut and RandomErasing
• Equalize
• AutoContrast
• GridMask
• SLIC Layer: Superpixel
• RandAug : PR
• AutoAugment
• AugMix
• Solarize : PR
• Shear : PR
• Sharpen
• RandomBrightness
• Color degeneration : PR
• Posterize
• Invert
• Simple Copy-Paste - Instance Segmentation
• Color Jitter : PR
• Grayscale : PR
• RGBShift : PR
• ChannelShuffle : PR
• Hue Value
• Saturation Value

[innat]

From keras.io

• Image preprocessing layers
• Image augmentation layers

[bhack]

One of these days we will probably need to care about how to handle classes in our "canned policies":

https://arxiv.org/abs/2204.03632.

[innat]

KerasCV 😵‍💫

[innat]

What an awesome Gif!!!!!! Would you share the script you used to create it?

haha, thanks. I made it on last week for fun. 😅
Though I didn't save the script I will share a new one.

[LukeWood]

What an awesome Gif!!!!!! Would you share the script you used to create it?

haha, thanks. I made it on last week for fun. 😅 Though I didn't save the script I will share a new one.

Dang! It’s a really cool visual. Perhaps a good addition to our marketing materials if CutMix + MixUp + fixed RandAugment are included 👀

[bhack]

Nice, but I suggest to change the image in marketing materials:

scikit-image/scikit-image#1855

[innat]

@LukeWood
Here are the scripts. COLAB.

@bhack
Yes, and data.astronaut() would be a nice choice. The official pytorch tutorials also use it.

[LukeWood]

Thank you @innat! I might produce one that uses CutMix and MixUp too and add it to our marketing :)

I'll be sure to reference you in the final version! Such a cool idea.

[bhack]

RegMixup: Mixup as a Regularizer Can Surprisingly Improve Accuracy and Out Distribution Robustness