2026/16更新 知乎专栏相关(细粒度)
https://zhuanlan.zhihu.com/fgva-cs
FGVC6 Workshop 共有十个挑战赛,每个都代表了细粒度视觉分类在某个细分领域的挑战
CVPR 2019 Cassava Disease Classification 比赛
2019 国内发表的一些好的领域的算法介绍
https://rrl360.com/boke/35114
1、2017鸟类的细粒度分类,粗到细的一个过程,加入了不同的损失
http://openaccess.thecvf.com/content_cvpr_2017/papers/Fu_Look_Closer_to_CVPR_2017_paper.pdf
代码还是有的
https://github.com/Jianlong-Fu/Recurrent-Attention-CNN/issues
https://pan.baidu.com/s/1Gx8ubL__Thc5LYTlX5hreA #网盘下载
2、2019京东蝶类精细度分类。
2019CVPR细粒度论文笔记《Destruction and Construction Learning for Fine-grained Image Recognition》
https://blog.csdn.net/zsx1713366249/article/details/92370490
2019-2020细腻度分类
1、A Systematic Evaluation:Fine-Grained CNN vs. Traditional CNN Classifiers 2020
主要对比分类模型,和细腻度模型方法,对细腻度数据的测试比对
通常的分类模型,densnet 比起最好的细腻度模型在细腻度数据上,效果更好一点
2 TIP2020 | 基于通道交互损失函数的细粒度图像分类方法
和上一篇正好测试结果互补,DCLresnet50,TASN ,MC_loss,Densenet161, NTS_net
最好。
https://github.com/dongliangchang/Mutual-Channel-Loss
3、BBN,BBN for Long-Tailed Visual Recognition(数据不平衡的训练)
特征提取模型和全连接模型分成两部分训练
文章首先指出了现在解决Long-Tail问题常用的re-balancing方法(Re-Weighting & Re-Sampling)虽然可以通过提升网络结构中classifier部分的性能,但是却损害了Representation部分(feature extractor backbone)学习的效果。如图1所示,虽然在Re-blancing之后,模型可以正确分类部分tail类的数据,但是却使得每个类的类内分布变得更加separable。
4、CVPR2020(ReID):Relation-Aware Global Attention for Person Re-identification
本文的创新是提出了一个Relation-Aware Global Attention (RGA) module,该module可以捕获全局结构信息用于注意力学习,精度提升几个点。(有实现)
https://github.com/microsoft/Relation-Aware-Global-Attention-Networks
对比比较常见的NL SE,明显提升
2017- 2019 细腻度分心代码和论文
ICCV2017 Dynamic Computational Time for Recurrent Attention Model (DT-RAM)
https://github.com/baidu-research/DT-RAM(Torch实现,ResNet的baseline很牛,CUB-200-2011数据集 84.5%,论文方法 86.0%)
http://openaccess.thecvf.com/content_ICCV_2017_workshops/papers/w18/Li_Dynamic_Computational_Time_ICCV_2017_paper.pdf
ECCV2018 Learning to Navigate for Fine-grained Classification
https://github.com/yangze0930/NTS-Net(Pytorch实现,CUB-200-2011数据集 87.6%)
http://openaccess.thecvf.com/content_ECCV_2018/papers/Ze_Yang_Learning_to_Navigate_ECCV_2018_paper.pdf
CVPR 2018 Learning a Discriminative Filter Bank Within a CNN for Fine-Grained Recognition
https://github.com/songdejia/DFL-CNN(Pytorch第三方实现,值得借鉴,但无法复现原文)
https://arxiv.org/abs/1611.09932(CUB-200-2011数据集 87.4%)
ECCV 2018 Pairwise Confusion for Fine-Grained Visual Classification
ECCV 2018 Multi-Attention Multi-Class Constraint for Fine-grained Image Recognition
ECCV 2018 Hierarchical Bilinear Pooling for Fine-Grained Visual Recognition
ECCV 2018 Grassmann Pooling as Compact Homogeneous Bilinear Pooling for Fine-Grained Visual Classification
CVPR 2019 Weakly Supervised Complementary Parts Models for Fine-Grained Image Classification From the Bottom Up
http://openaccess.thecvf.com/content_CVPR_2019/papers/Ge_Weakly_Supervised_Complementary_Parts_Models_for_Fine-Grained_Image_Classification_From_CVPR_2019_paper.pdf(CUB-200-2011数据集 90.4%,用了比较多的策略:+ Stacked LSTM + Multi-Loss,Baseline 82.8%)
CVPR 2019 Looking for the Devil in the Details: Learning Trilinear Attention Sampling Network for Fine-Grained Image Recognition
https://github.com/researchmm/tasn(MXNet实现,CUB-200-2011数据集 87.0%)
http://openaccess.thecvf.com/content_CVPR_2019/papers/Zheng_Looking_for_the_Devil_in_the_Details_Learning_Trilinear_Attention_CVPR_2019_paper.pdf
CVPR 2019 Destruction and Construction Learning for Fine-Grained Image Recognition
https://github.com/JDAI-CV/DCL(Pytorch实现,CUB-200-2011数据集 87.8%)
http://openaccess.thecvf.com/content_CVPR_2019/papers/Chen_Destruction_and_Construction_Learning_for_Fine-Grained_Image_Recognition_CVPR_2019_paper.pdf
ArXiv 2019 See Better Before Looking Closer: Weakly Supervised Data Augmentation Network for Fine-Grained Visual Classification
https://github.com/tau-yihouxiang/WS_DAN(Tensorflow,官方实现)
https://github.com/GuYuc/WS-DAN.PyTorch(Pytorch,第三方实现)
https://arxiv.org/abs/1901.09891
https://zhuanlan.zhihu.com/p/94799831
2018 Learning a Discriminative Filter Bank within a CNN for Fine-grained Recognition,细粒度图像识别
来源,ssd 启发
https://blog.csdn.net/KongCDY/article/details/100009420