LatentHuman: Shape-and-Pose Disentangled Latent Representation for for Human Bodies abs: https://t.co/gsZdXH1Sik project page: https://t.co/oenzQJPC0d

SketchEdit: Mask-Free Local Image Manipulation with Partial Sketches abs: https://t.co/bywtQTZUHF project page: https://t.co/6hkNEQGmZU github: https://t.co/hGvuDkJT7X

MultiPath++: Efficient Information Fusion and Trajectory Aggregation for Behavior Prediction abs: https://t.co/rozuYsDMCU model achieves sota performance on the Argoverse Motion Forecasting Competition and the Waymo Open Dataset Motion Prediction Challenge

Text Mining Drug/Chemical-Protein Interactions using an Ensemble of BERT and T5 Based Models abs: https://t.co/Gx7VMhYJcV

train a single hypernetwork to learn how to refine the generator weights with respect to a desired target image. By learning this mapping, HyperStyle efficiently predicts the desired generator weights in less than 2 seconds per image

HyperStyle: StyleGAN Inversion with HyperNetworks for Real Image Editing abs: https://t.co/IkXFzcrKpZ project page: https://t.co/lkOcMjeO3o HyperStyle yields reconstructions comparable to those of optimization techniques with the near real-time inference capabilities of encoders

AdaViT: Adaptive Vision Transformers for Efficient Image Recognition abs: https://t.co/JkGgzi64CW experiments on ImageNet, method obtains more than 2× improvement on efficiency compared to sota vision transformers with 0.8% drop of accuracy

ATS: Adaptive Token Sampling For Efficient Vision Transformers abs: https://t.co/pqx557Ebe7 evaluations show that the proposed module improves the sota by reducing the computational cost (GFLOPs) by 37% while preserving the accuracy

Donut 🍩 : Document Understanding Transformer without OCR abs: https://t.co/A644UXgUuG achieves sota performance on various document understanding tasks in public benchmark datasets and private industrial service datasets

Shunted Self-Attention via Multi-Scale Token Aggregation abs: https://t.co/iJV2c2noIa github: https://t.co/AX0cQRTQAH

Pyramid Adversarial Training Improves ViT Performance abs: https://t.co/oaxB6Q99R2 new sota for ImageNet-C (41.4 mCE), ImageNetR (53.92%), and ImageNet-Sketch (41.04%) without extra data, using only the ViT-B/16 backbone and pyramid adversarial training

DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation abs: https://t.co/IePs79GeHI github: https://t.co/sI0P4olATR improves the sota performance by 10.8 mIoU for GTA→Cityscapes and 5.4 mIoU for Synthia→Cityscapes

NeuSample: Neural Sample Field for Efficient View Synthesis abs: https://t.co/ND0SNmZgh8 project page: https://t.co/ym5NZplUDM

FENeRF: Face Editing in Neural Radiance Fields abs: https://t.co/KhTLJaN9Ws a 3D-aware generator that can produce view-consistent and locally-editable portrait images

Hallucinated Neural Radiance Fields in the Wild abs: https://t.co/ojyyTRI8oV project page: https://t.co/2wGcJUYipD Neural Body can reconstruct a moving human from a monocular video

NeRFReN: Neural Radiance Fields with Reflections abs: https://t.co/GSrJpBWjqP project page: https://t.co/fBqtVcrZ23 Experiments on self-captured scenes show method achieves high-quality NVS and physically sound depth estimation results while enabling scene editing applications

Diffusion Autoencoders: Toward a Meaningful and Decodable Representation abs: https://t.co/YMxRdOM0CW project page: https://t.co/eeN565mX7X can encode any image into a two-part latent code, first part is semantically meaningful and linear, second part captures stochastic details

CRIS: CLIP-Driven Referring Image Segmentation abs: https://t.co/fdxUBm0WNB experimental results on three benchmark datasets demonstrate that proposed framework significantly outperforms sota performance without any post-processing

Improving Zero-shot Generalization in Offline Reinforcement Learning using Generalized Similarity Functions abs: https://t.co/5B163MSLHN