来源:机器之心
编辑:蛋酱
虽然世界仍在从新冠疫情的破坏中复苏,人们无法向从前那样时常线下相聚、共同探讨交流关于学术领域的最新问题,但AI研究也没有停下跃进的步伐。
转眼就是2021年底了,一年就这么就过去了,时光好像被偷走一样。细细数来,你今年读了多少论文?
一名加拿大博主Louis Bouchard以发布时间为顺序,整理出了近40篇2021年不可错过的优秀论文。整体来看,合集中的论文偏重计算机视觉方向。
在这个15分钟左右的视频中,你可以快速浏览这些论文:
以下是每篇论文的详细信息:
1、DALL·E: Zero-Shot Text-to-Image Generation from OpenAI
论文链接:https://arxiv.org/pdf/2102.12092.pdf
代码地址:https://github.com/openai/DALL-E
视频解读:https://youtu.be/DJToDLBPovg
2、VOGUE: Try-On by StyleGAN Interpolation Optimization
论文链接:https://vogue-try-on.github.io/static_files/resources/VOGUE-virtual-try-on.pdf
视频解读:https://youtu.be/i4MnLJGZbaM
3、Taming Transformers for High-Resolution Image Synthesis
论文链接:https://compvis.github.io/taming-transformers/
代码地址:https://github.com/CompVis/taming-transformers
视频解读:https://youtu.be/JfUTd8fjtX8
4、Thinking Fast And Slow in AI
论文链接:https://arxiv.org/abs/2010.06002
视频解读:https://youtu.be/3nvAaVSQxs4
5、Automatic detection and quantification of floating marine macro-litter in aerial images
论文链接:https://doi.org/10.1016/j.envpol.2021.116490
代码地址:https://github.com/amonleong/MARLIT
视频解读:https://youtu.be/2dTSsdW0WYI
6、ShaRF: Shape-conditioned Radiance Fields from a Single View
论文链接:https://arxiv.org/abs/2102.08860
代码地址:http://www.krematas.com/sharf/index.html
视频解读:https://youtu.be/gHkkrNMlGNg
7、Generative Adversarial Transformers
论文链接:https://arxiv.org/pdf/2103.01209.pdf
代码地址:https://github.com/dorarad/gansformer
视频解读:https://youtu.be/HO-_t0UArd4
8、We Asked Artificial Intelligence to Create Dating Profiles. Would You Swipe Right?
论文链接:https://studyonline.unsw.edu.au/blog/ai-generated-dating-profile
代码地址:https://colab.research.google.com/drive/1VLG8e7YSEwypxU-noRNhsv5dW4NfTGce#forceEdit=true&sandboxMode=true&scrollTo=aeXshJM-Cuaf
视频解读:https://youtu.be/IoRH5u13P-4
9、Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
论文链接:https://arxiv.org/abs/2103.14030v2
代码地址:https://github.com/microsoft/Swin-Transformer
视频解读:https://youtu.be/QcCJJOLCeJQ
10、IMAGE GANS MEET DIFFERENTIABLE RENDERING FOR INVERSE GRAPHICS AND INTERPRETABLE 3D NEURAL RENDERING
论文链接:https://arxiv.org/pdf/2010.09125.pdf
视频解读:https://youtu.be/dvjwRBZ3Hnw
11、Deep nets: What have they ever done for vision?
论文链接:https://arxiv.org/abs/1805.04025
视频解读:https://youtu.be/GhPDNzAVNDk
12、Infinite Nature: Perpetual View Generation of Natural Scenes from a Single Image
论文链接:https://arxiv.org/pdf/2012.09855.pdf
代码地址:https://github.com/google-research/google-research/tree/master/infinite_nature
视频解读:https://youtu.be/NIOt1HLV_Mo
在线试用:https://colab.research.google.com/github/google-research/google-research/blob/master/infinite_nature/infinite_nature_demo.ipynb#scrollTo=sCuRX1liUEVM
13、Portable, Self-Contained Neuroprosthetic Hand with Deep Learning-Based Finger Control
论文链接:https://arxiv.org/abs/2103.13452
视频解读:https://youtu.be/wNBrCRzlbVw
14、Total Relighting: Learning to Relight Portraits for Background Replacement
论文链接:https://augmentedperception.github.io/total_relighting/total_relighting_paper.pdf
视频解读:https://youtu.be/rVP2tcF_yRI
15、LASR: Learning Articulated Shape Reconstruction from a Monocular Video
论文链接:https://openaccess.thecvf.com/content/CVPR2021/papers/Yang_LASR_Learning_Articulated_Shape_Reconstruction_From_a_Monocular_Video_CVPR_2021_paper.pdf
代码地址:https://github.com/google/lasr
视频解读:https://youtu.be/lac7wqjS-8E
16、Enhancing Photorealism Enhancement
论文链接:http://vladlen.info/papers/EPE.pdf
代码地址:https://github.com/isl-org/PhotorealismEnhancement
视频解读:https://youtu.be/3rYosbwXm1w
17、DefakeHop: A Light-Weight High-Performance Deepfake Detector
论文链接:https://arxiv.org/abs/2103.06929
视频解读:https://youtu.be/YMir8sRWRos
18、High-Resolution Photorealistic Image Translation in Real-Time: A Laplacian Pyramid Translation Network
论文链接:https://arxiv.org/pdf/2105.09188.pdf
代码地址:https://github.com/csjliang/LPTN
视频解读:https://youtu.be/X7WzlAyUGPo
19、Barbershop: GAN-based Image Compositing using Segmentation Masks
论文链接:https://arxiv.org/pdf/2106.01505.pdf
代码地址:https://github.com/ZPdesu/Barbershop
视频解读:https://youtu.be/HtqYMvBVJD8
20、TextStyleBrush: Transfer of text aesthetics from a single example
论文链接:https://arxiv.org/abs/2106.08385
代码地址:https://github.com/facebookresearch/IMGUR5K-Handwriting-Dataset?fbclid=IwAR0pRAxhf8Vg-5H3fA0BEaRrMeD21HfoCJ-so8V0qmWK7Ub21dvy_jqgiVo
视频解读:https://youtu.be/hhAri5fl-XI
21、Animating Pictures with Eulerian Motion Fields
论文链接:https://arxiv.org/abs/2011.15128
代码地址:https://eulerian.cs.washington.edu/
视频解读:https://youtu.be/KgTa2r7d0I0
22、CVPR 2021 Best Paper Award: GIRAFFE - Controllable Image Generation
论文链接:http://www.cvlibs.net/publications/Niemeyer2021CVPR.pdf
代码地址:https://github.com/autonomousvision/giraffe
视频解读:https://youtu.be/JIJkURAkCxM
23、GitHub Copilot & Codex: Evaluating Large Language Models Trained on Code
论文链接:https://arxiv.org/pdf/2107.03374.pdf
代码地址:https://copilot.github.com/
视频解读:https://youtu.be/az3oVVkTFB8
24、Apple: Recognizing People in Photos Through Private On-Device Machine Learning
论文链接:https://machinelearning.apple.com/research/recognizing-people-photos
视频解读:https://youtu.be/LIV-M-gFRFA
25、Image Synthesis and Editing with Stochastic Differential Equations
论文链接:https://arxiv.org/pdf/2108.01073.pdf
代码地址:https://github.com/ermongroup/SDEdit
视频解读:https://youtu.be/xoEkSWJSm1k
https://colab.research.google.com/drive/1KkLS53PndXKQpPlS1iK-k1nRQYmlb4aO?usp=sharing
26、Sketch Your Own GAN
论文链接:https://arxiv.org/abs/2108.02774
代码地址:https://github.com/PeterWang512/GANSketching
视频解读:https://youtu.be/vz_wEQkTLk0
27、Tesla's Autopilot Explained
在今年8月的特斯拉AI日上,特斯拉AI总监Andrej Karpathy和其他人展示了特斯拉是如何通过八个摄像头采集图像,打造了基于视觉的自动驾驶系统。
视频解读:https://youtu.be/DTHqgDqkIRw
28、Styleclip: Text-driven manipulation of StyleGAN imagery
论文链接:https://arxiv.org/abs/2103.17249
代码地址:https://github.com/orpatashnik/StyleCLIP
视频解读:https://youtu.be/RAXrwPskNso
https://colab.research.google.com/github/orpatashnik/StyleCLIP/blob/main/notebooks/StyleCLIP_global.ipynb
29、TimeLens: Event-based Video Frame Interpolation
论文链接:http://rpg.ifi.uzh.ch/docs/CVPR21_Gehrig.pdf
代码地址:https://github.com/uzh-rpg/rpg_timelens
视频解读:https://youtu.be/HWA0yVXYRlk
30、Diverse Generation from a Single Video Made Possible
论文链接:https://arxiv.org/abs/2109.08591
代码地址:https://nivha.github.io/vgpnn/
视频解读:https://youtu.be/Uy8yKPEi1dg
31、Skillful Precipitation Nowcasting using Deep Generative Models of Radar
论文链接:https://www.nature.com/articles/s41586-021-03854-z
代码地址:https://github.com/deepmind/deepmind-research/tree/master/nowcasting
视频解读:https://youtu.be/dlSIq64psEY
32、The Cocktail Fork Problem: Three-Stem Audio Separation for Real-World Soundtracks
论文链接:https://arxiv.org/pdf/2110.09958.pdf
代码地址:https://cocktail-fork.github.io/
视频解读:https://youtu.be/Rpxufqt5r6I
33、ADOP: Approximate Differentiable One-Pixel Point Rendering
论文链接:https://arxiv.org/pdf/2110.06635.pdf
代码地址:https://github.com/darglein/ADOP
视频解读:https://youtu.be/Jfph7Vld_Nw
34、(Style)CLIPDraw: Coupling Content and Style in Text-to-Drawing Synthesis
CLIPDraw论文链接:https://arxiv.org/abs/2106.14843
在线试用:https://colab.research.google.com/github/kvfrans/clipdraw/blob/main/clipdraw.ipynb
StyleCLIPDraw论文链接:https://arxiv.org/abs/2111.03133
在线试用:https://colab.research.google.com/github/pschaldenbrand/StyleCLIPDraw/blob/master/Style_ClipDraw.ipynb
视频解读:https://youtu.be/5xzcIzHm8Wo
35、SwinIR: Image restoration using swin transformer
论文链接:https://arxiv.org/abs/2108.10257
代码地址:https://github.com/JingyunLiang/SwinIR
视频解读:https://youtu.be/GFm3RfrtDoU
https://replicate.ai/jingyunliang/swinir
36、EditGAN: High-Precision Semantic Image Editing
论文链接:https://arxiv.org/abs/2111.03186
代码地址:https://nv-tlabs.github.io/editGAN/
视频解读:https://youtu.be/bus4OGyMQec
37、CityNeRF: Building NeRF at City Scale
论文链接:https://arxiv.org/pdf/2112.05504.pdf
代码地址:https://city-super.github.io/citynerf/
视频解读:https://youtu.be/swfx0bJMIlY
38、ClipCap: CLIP Prefix for Image Captioning
论文链接:https://arxiv.org/abs/2111.09734
代码地址:https://github.com/rmokady/CLIP_prefix_caption
视频解读:https://youtu.be/VQDrmuccWDo
在线试用:https://colab.research.google.com/drive/1tuoAC5F4sC7qid56Z0ap-stR3rwdk0ZV?usp=sharing
当然,博主在整理的过程中也不能保证完美。经网友提醒,这里可以手动添加一项突破性研究:「AlphaFold」。
去年,谷歌旗下人工智能技术公司 DeepMind 宣布深度学习算法「Alphafold」破解了出现五十年之久的蛋白质分子折叠问题。2021年7月,AlphaFold 的论文正式发表在《Nature》杂志上。
论文链接:https://www.nature.com/articles/s41586-021-03819-2
这项研究被评为Nature年度技术突破,Alphafold 的缔造者之一 John Jumper 也被评为《Nature》2021 年度十大科学人物。DeepMind也已经将他们的预测结果免费开放给公众。
对于你来说,2021年最令人印象深刻的论文又是哪篇呢?
原文链接:https://www.louisbouchard.ai/2021-ai-papers-review/
未来智能实验室的主要工作包括:建立AI智能系统智商评测体系,开展世界人工智能智商评测;开展互联网(城市)大脑研究计划,构建互联网(城市)大脑技术和企业图谱,为提升企业,行业与城市的智能水平服务。每日推荐范围未来科技发展趋势的学习型文章。目前线上平台已收藏上千篇精华前沿科技文章和报告。
如果您对实验室的研究感兴趣,欢迎加入未来智能实验室线上平台。扫描以下二维码或点击本文左下角“阅读原文”