21年 46篇神经架构搜索(NAS) ICCV CVPR Survey 笔记 (21-46持续更新)

目录

21. AutoSpace: Neural Architecture Search with Less Human Interference(ICCV)

22. CM-NAS: Cross-Modality Neural Architecture Search for Visible-Infrared Person Re-Identifification

23. Joint-DetNAS: Upgrade Your Detector with NAS, Pruning and Dynamic Distillation

24. NetAdaptV2: Effificient Neural Architecture Search with Fast Super-Network Training and Architecture Optimization

25. NPAS: A Compiler-aware Framework of Unifified Network Pruning and Architecture Search for Beyond Real-Time Mobile Acceleration

26. OPANAS: One-Shot Path Aggregation Network Architecture Search for Object Detection

27. Pi-NAS: Improving Neural Architecture Search by Reducing Supernet Training Consistency Shift

28. Prioritized Architecture Sampling with Monto-Carlo Tree Search

29.TransNAS-Bench-101: Improving transferability and Generalizability of Cross-Task Neural Architecture Search

30. ViPNAS: Effificient Video Pose Estimation via Neural Architecture Search

31. Compatibility-aware Heterogeneous Visual Search

32. DiNTS: Differentiable Neural Network Topology Search for 3D Medical Image Segmentation

33. Distilling Optimal Neural Networks: Rapid Search in Diverse Spaces

34. FP-NAS: Fast Probabilistic Neural Architecture Search

35. Homogeneous Architecture Augmentation for Neural Predictor

36.Landmark Regularization: Ranking Guided Super-Net Training in Neural Architecture Search

37. Learning Latent Architectural Distribution in Differentiable Neural Architecture Search via Variational Information Maximization

38. NASOA: Towards Faster Task-oriented Online Fine-tuning with a Zoo of Models

39. NAS-OoD: Neural Architecture Search for Out-of-Distribution Generalization

40.Neural Architecture Search for Joint Human Parsing and Pose Estimation

41. iNAS: Integral NAS for Device-Aware Salient Object Detection

42.Not All Operations Contribute Equally: Hierarchical Operation-adaptive Predictor for Neural Architecture Search

43.One-Shot Neural Ensemble Architecture Search by Diversity-Guided Search Space Shrinking

44. RANK-NOSH: Effificient Predictor-Based Architecture Search via Non-Uniform Successive Halving

45.Rethinking Graph Neural Architecture Search from Message-passing

46.Searching by Generating: Flexible and Effificient One-Shot NAS with Architecture Generator

---

21. AutoSpace: Neural Architecture Search with Less Human Interference(ICCV)

Aim:	新算法独立自主生成搜索空间，NAS目前仍然需要专家来设计搜索空间。
Abstract：	目前的算法仍然需要专家来设计搜索空间，本文考虑了自动化设计一个搜索空间。这将会面临两个问题：         (1)搜索空间爆炸性的复杂度         ( 2)评估不同空间质量的昂贵计算成本         为了解决这些问题，提出可微的适应度评分函数，有效的评估cell的性能和参考架构加速进化过程，避免陷入次优解的方案。         框架通用，兼容而外的计算约束，这使得学习适合不同计算预算的专门搜索空间成为可行的。
Conclusion:	移动模式下(model<500M) ImageNet 精确度 77.8%
Methods:	新的可微进化框架、适应度评价函数、额外约束条件
Keyresults:	可微搜索、进化算法
Code:	GitHub - zhoudaquan/AutoSpace   只公布了一部分代码

 

22. CM-NAS: Cross-Modality Neural Architecture Search for Visible-Infrared Person Re-Identifification

这篇不像常见的NAS算法。

Aim:	VI-ReID 可见红外线人员重新识别：目的匹配跨模态的行人图像，打破单模态的reID局限性。
Abstract：	现有的人工设计各种双流架构，分别学习特定模态的和模态可共享的表示. 确定适当分离BN层，是促进跨空间性匹配的关键。 新算法由一个面向bn的搜索空间组成，在其中可以通过跨模态任务来完成标准的优化
Conclusion:	在两个基准测试中都优于最先进的方法 改进了SYSU-MM01上的6.70%/6.13%和RegDB上的12.17%/11.23%改进了Rank-1/mAP
Methods:	Cross-Modality Neural Architecture Search (CM- NAS).
Keyresults:	SYSU-MM01 、RegDB、Visible-Infrared person re-identifification (VI-ReID)
Code:	GitHub - JDAI-CV/CM-NAS: CM-NAS: Cross-Modality Neural Architecture Search for Visible-Infrared Person Re-Identification (ICCV2021)  代码很详细

 

 

23. Joint-DetNAS: Upgrade Your Detector with NAS, Pruning and Dynamic Distillation

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	无代码

 

24. NetAdaptV2: Effificient Neural Architecture Search with Fast Super-Network Training and Architecture Optimization

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	NetAdapt Project  项目地址在这，只公布了V1，V2没有公布

25. NPAS: A Compiler-aware Framework of Unifified Network Pruning and Architecture Search for Beyond Real-Time Mobile Acceleration

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	无代码

26. OPANAS: One-Shot Path Aggregation Network Architecture Search for Object Detection

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	VDIGPKU/OPANAS: The official code for OPANAS: One-Shot Path Aggregation Network Architecture Search for Object Detection (CVPR 2021) (github.com)

27. Pi-NAS: Improving Neural Architecture Search by Reducing Supernet Training Consistency Shift

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	https://github.com/Ernie1/Pi-NAS

28. Prioritized Architecture Sampling with Monto-Carlo Tree Search

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	GitHub - xiusu/NAS-Bench-Macro: NAS Benchmark in "Prioritized Architecture Sampling with Monto-Carlo Tree Search", CVPR2021

29.TransNAS-Bench-101: Improving transferability and Generalizability of Cross-Task Neural Architecture Search

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	Index of /dataset/TransNAS-Bench-101/ 相关项目地址 Vega

30. ViPNAS: Effificient Video Pose Estimation via Neural Architecture Search

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	论文中有一些相关代码，但是并没有给全。

31. Compatibility-aware Heterogeneous Visual Search

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	没开源

32. DiNTS: Differentiable Neural Network Topology Search for 3D Medical Image Segmentation

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	没开源

33. Distilling Optimal Neural Networks: Rapid Search in Diverse Spaces

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	没开源

34. FP-NAS: Fast Probabilistic Neural Architecture Search

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	没开源

35. Homogeneous Architecture Augmentation for Neural Predictor(ICCV)

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	GitHub - lyq998/HAAP: Homogeneous Architecture Augmentation for Neural Predictor

36.Landmark Regularization: Ranking Guided Super-Net Training in Neural Architecture Search

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	GitHub - kcyu2014/nas-landmarkreg: [CVPR2021] Code for Landmark Regularization: Ranking Guided Super-Net Training in Neural Architecture Search

37. Learning Latent Architectural Distribution in Differentiable Neural Architecture Search via Variational Information Maximization

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	没开源

38. NASOA: Towards Faster Task-oriented Online Fine-tuning with a Zoo of Models

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	提供训练模型 https://github.com/NAS-OA/NASOA

39. NAS-OoD: Neural Architecture Search for Out-of-Distribution Generalization

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	没开源

40.Neural Architecture Search for Joint Human Parsing and Pose Estimation

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	https://github.com/GuHuangAI/NPP

41. iNAS: Integral NAS for Device-Aware Salient Object Detection

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	主页iNAS: Integral NAS for Device-Aware Salient Object Detection – 程明明个人主页 代码GitHub - guyuchao/iNAS: Open Source Neural Architecture Search Toolbox for Device-aware Image Dense Prediction & Official implementation of ICCV2021 "iNAS: Integral NAS for Device-Aware Salient Object Detection"

42.Not All Operations Contribute Equally: Hierarchical Operation-adaptive Predictor for Neural Architecture Search

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	没代码

43.One-Shot Neural Ensemble Architecture Search by Diversity-Guided Search Space Shrinking

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	https://github.com/researchmm/NEAS

44. RANK-NOSH: Effificient Predictor-Based Architecture Search via Non-Uniform Successive Halving

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	没开源

45.Rethinking Graph Neural Architecture Search from Message-passing

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	https://github.com/phython96/GNAS-MP

46.Searching by Generating: Flexible and Effificient One-Shot NAS with Architecture Generator

Aim:
Abstract：
Conclusion:
Methods:
Keyresults:
Code:	https://github.com/eric8607242/SGNAS