MODALS: Modality-agnostic Automated Data Augmentation in the Latent Space
This video explains a technique for domain agnostic data augmentation accepted into the ICLR 2021 conference. I am really excited about this technique to bring the success of data augmentation in Computer Vision to more applications!
Content Links:
Paper: https://openreview.net/forum?id=XjYgR...
Robert Luxemburg's video on StyleGAN2 interpolation: • StyleGAN2 Interpolation Loop
Population-Based Augmentation: https://arxiv.org/abs/1905.05393
Cutout: https://arxiv.org/pdf/1708.04552.pdf
Easy Data Augmentation: https://arxiv.org/pdf/1901.11196.pdf
TensorFlow Data Augmentation: https://www.tensorflow.org/tutorials/...
StyleGAN: https://arxiv.org/abs/1912.04958
Dataset Augmentation in Feature Space: https://arxiv.org/pdf/1702.05538.pdf
Population-Based Training of Neural Networks: https://arxiv.org/pdf/1711.09846.pdf
BigGAN: https://arxiv.org/pdf/1809.11096.pdf
SimCLR: https://arxiv.org/pdf/2002.05709.pdf
Generative Teaching Networks: https://arxiv.org/pdf/1912.07768.pdf
Kaggle Competitions: https://www.kaggle.com/competitions
Chapters:
0:00 Beginning
1:29 What is Data Augmentation?
4:09 Latent Space Augmentation
6:31 Other tools available
7:54 Algorithm Deep Dive
17:27 Initial Results
19:21 Population-based Augmentation
22:10 Results with PBA and Ablations
23:14 Kaggle Competitions (not just images…)
23:43 Contrastive Learning outside of Vision
24:02 Outer-inner loop ideas for Domain-Agnostic Data Aug
Thanks for watching! Please Subscribe for more DL videos!