CVPR-2022 Paper Digest

Every paper, ordered by impact or potential impact. About CVPR 2022

IDPaperAuthors
1High-Resolution Image Synthesis With Latent Diffusion Models
IF:10   Related Papers   Related Patents   Related Grants   Related Venues   Related Experts   Related Code   View    Save
Highlight: By introducing cross-attention layers into the model architecture, we turn diffusion models into powerful and flexible generators for general conditioning inputs such as text or bounding boxes and high-resolution synthesis becomes possible in a convolutional manner.
Robin Rombach; Andreas Blattmann; Dominik Lorenz et al.
2Blended Diffusion for Text-Driven Editing of Natural Images
IF:9   Related Papers   Related Patents   Related Grants   Related Venues   Related Experts   Related Code   View    Save
Highlight: In this paper, we introduce the first solution for performing local (region-based) edits in generic natural images, based on a natural language description along with an ROI mask.
Omri Avrahami; Dani Lischinski; Ohad Fried;
3Masked Autoencoders Are Scalable Vision Learners
IF:9   Related Papers   Related Patents   Related Grants   Related Venues   Related Experts   Related Code   View    Save
Highlight: This paper shows that masked autoencoders (MAE) are scalable self-supervised learners for computer vision.
Kaiming He; Xinlei Chen; Saining Xie et al.
4Masked-Attention Mask Transformer for Universal Image Segmentation
IF:9   Related Papers   Related Patents   Related Grants   Related Venues   Related Experts   Related Code   View    Save
Highlight: We present Masked-attention Mask Transformer (Mask2Former), a new architecture capable of addressing any image segmentation task (panoptic, instance or semantic).
Bowen Cheng; Ishan Misra; Alexander G. Schwing et al.
5Depth-Supervised NeRF: Fewer Views and Faster Training for Free
IF:9   Related Papers   Related Patents   Related Grants   Related Venues   Related Experts   Related Code   View    Save
Highlight: One potential reason is that standard volumetric rendering does not enforce the constraint that most of a scene's geometry consist of empty space and opaque surfaces. We formalize the above assumption through DS-NeRF (Depth-supervised Neural Radiance Fields), a loss for learning radiance fields that takes advantage of readily-available depth supervision.
Kangle Deng; Andrew Liu; Jun-Yan Zhu et al.

About CVPR 2022