Photoshop detector AI called FALDetector aims to detect facial edits that warps the faces to make chin or jaw looks thinner, or the forehead smaller. However, it is not really accurate and consistent as it look at the individual pixels too much, as different resolution of the same person can result in different predictions with this AI model.

Transferable Interactiveness Networks (TIN) for Human-Object Interaction (HOI) Detection

Create Character Animation with small amount of data by using Generative Adversarial Network (GAN)

Github:

https://github.com/tohinz/CharacterGAN

Paper:

https://arxiv.org/pdf/2102.03141.pdf

You can use also interactive GUI to easily repose a given character based on keypoints.

End-to-end speech recognition toolkit (based on Pytorch)
What's make it different from other ASR toolkits:
* Production first and production ready
* Unified solution for streaming and non-streaming ASR
* Portable runtime
* Light weight

"Unbiased Teacher for Semi-Supervised Object Detection" from Facebook researchers

Detectron2 is a ground-up rewrite of Detectron that started with maskrcnn-benchmark.
The platform is now implemented in PyTorch.

CVPR 2021 paper

Modular Interactive Video Object Segmentation: Interaction-to-Mask, Propagation and Difference-Aware Fusion (MiVOS)

YolactEdge Real time Instance Segmentation on the Edge
https://www.youtube.com/watch?v=pMDwXkIerw8

Github:

https://github.com/ai-coodinator/yolact_edge

YolactEdge, competitive instance segmentation approach that runs on small edge devices at real-time speeds. Specifically, YolactEdge runs at up to 30.8 FPS on a Jetson AGX Xavier (and 172.7 FPS on an RTX 2080 Ti) with a ResNet-101 backbone on 550x550 resolution images.

Timers and Such: A Practical Benchmark for Spoken Language Understanding with Numbers

End-to-end pipeline for Spoken Language Understanding (SLU)