SimCLR

Abstract
SimCLR is a simple framework for contrastive learning of visual representations. We simplify recently proposed contrastive self-supervised learning algorithms without requiring specialized architectures or a memory bank. In order to understand what enables the contrastive prediction tasks to learn useful representations, we systematically study the major components of our framework.
We show that (1) composition of data augmentations plays a critical role in defining effective predictive tasks, (2) introducing a learnable nonlinear transformation between the representation and the contrastive loss substantially improves the quality of the learned representations, and (3) contrastive learning benefits from larger batch sizes and more training steps compared to supervised learning. By combining these findings, we are able to considerably outperform previous methods for self-supervised and semi-supervised learning on ImageNet. A linear classifier trained on self-supervised representations learned by SimCLR achieves 76.5% top-1 accuracy, which is a 7% relative improvement over previous state-of-the-art, matching the performance of a supervised ResNet-50. When fine-tuned on only 1% of the labels, we achieve 85.8% top-5 accuracy, outperforming AlexNet with 100X fewer labels.

* Available in pytorch-lightning-bolts with a couple of code lines
* Paper
* Colab implementation
* Explanation and implementation details in a series of short videos
#self_supervised

NYC Deep Learning Course: Structured Prediction

Final lecture of the course on deep learning led by Yann LeCun. It covers structured prediction, energy-based factor graphs, and graph transformer networks.

https://poloclub.github.io/ganlab/
Милая интерактивная демка с объяснением ганов на tf.js.

Flame plugin for blender for random face expression and random face generation
#morphable_model

Unity launches open projects — creating game in open source where anybody can contribute (programmers, artists, crocodiles).
https://youtu.be/jrQimv_7gcc

⭐️ Github repository for Project #1
⭐️ Roadmap for Project #1

​​NVidia released a technology to change face alignment on video

Nvidia has unveiled AI face-alignment that means you're always looking at the camera during video calls. Its new Maxine platform uses GANs to reconstruct the unseen parts of your head — just like a deepfake.

Link: https://www.theverge.com/2020/10/5/21502003/nvidia-ai-videoconferencing-maxine-platform-face-gaze-alignment-gans-compression-resolution

#NVidia #deepfake #GAN

Найдено у CG_Vines.
Оригинальная линка на реддит.

https://ebsynth.com/
Если получится поиграть с приложением на макоси, дам знать.
Саму идею мы уже видели, а вот приложение еще нет.
Tutorial: https://youtu.be/0RLtHuu5jV4

Попробовала сделать короткий видос с одним кейфреймом. Из важных особенностей — контуры на обрисованном видео и исходном видео должны совпадать. Нельзя пририсовать рожки, если их не было на видео или сделать нож из шаурмы.

картинка

АААА!!!
Этот чувак делает просто отличные ревью статей.
Это очень забавно.
https://youtu.be/TrdevFK_am4