整理：AI算法与图像处理
CVPR2022论文和代码整理：https://github.com/DWCTOD/CVPR2022-Papers-with-Code-Demo
ECCV2022论文和代码整理：https://github.com/DWCTOD/ECCV2022-Papers-with-Code-Demo
最新成果demo展示：
标题：
eDiffi: Text-to-Image Diffusion Models with an Ensemble of Expert Denoisers
论文：https://arxiv.org/pdf/2211.01324.pdf
主页：https://deepimagination.cc/eDiffi/
官方表述：Compared to the open-source text to image methods (Stable diffusion) and (DALL-E2), our model conistently leads to improved synthesis quality.
我们提出了eDiffi，一种用于合成给定文本的图像的扩散模型。基于扩散模型在不同采样阶段的行为不同的经验观察，我们建议训练一组专家去噪网络，每个网络专门针对特定的噪声区间。我们的模型以T5文本嵌入、CLIP图像嵌入和CLIP文本嵌入为条件。我们的方法可以生成与任何输入文本提示相对应的照片级真实感图像。除了文本到图像合成，我们还提供了两个额外的功能-（1）样式转换，它使我们能够使用参考样式图像控制生成的样本的样式；（2）“用文字绘制”-一个用户可以通过在画布上绘制分割图来生成图像的应用程序，这对于在脑海中绘制所需图像非常方便。
最新论文整理
ECCV2022
Updated on : 7 Nov 2022
total number : 3
Soft Masking for Cost-Constrained Channel Pruning
- 论文/Paper: http://arxiv.org/pdf/2211.02206
- 代码/Code: https://github.com/nvlabs/smcp
Large Scale Real-World Multi-Person Tracking
- 论文/Paper: http://arxiv.org/pdf/2211.02175
- 代码/Code: None
Abstract Images Have Different Levels of Retrievability Per Reverse Image Search Engine
- 论文/Paper: http://arxiv.org/pdf/2211.02115
- 代码/Code: None
CVPR2022
NeurIPS
Updated on : 7 Nov 2022
total number : 5
SelecMix: Debiased Learning by Contradicting-pair Sampling
- 论文/Paper: http://arxiv.org/pdf/2211.02291
- 代码/Code: None
Unsupervised Visual Representation Learning via Mutual Information Regularized Assignment
- 论文/Paper: http://arxiv.org/pdf/2211.02284
- 代码/Code: None
Zero-shot Video Moment Retrieval With Off-the-Shelf Models
- 论文/Paper: http://arxiv.org/pdf/2211.02178
- 代码/Code: None
Scaling Multimodal Pre-Training via Cross-Modality Gradient Harmonization
- 论文/Paper: http://arxiv.org/pdf/2211.02077
- 代码/Code: None
A Theoretical Study on Solving Continual Learning
- 论文/Paper: http://arxiv.org/pdf/2211.02633
- 代码/Code: https://github.com/k-gyuhak/wptp
