python main_ce.py --batch_size 1024 \ --learning_rate 0.8 \ --cosine --syncBN \ (2) Supervised Contrastive Learning Pretraining stage: python main_supcon.py --batch_size 1024 \ --learning_rate 0.5 \ --temp 0.1 \ --cosine
raymin0223/self-contrastive-learning Star19 Self-Contrastive Learning: Single-viewed Supervised Contrastive Framework using Sub-network (AAAI 2023) pytorchcontrastive-learningsupervised-contrastive-learningmulti-exit-architecturessingle-viewed-contrastive
LTH14/targeted-supcon: A PyTorch implementation of the paper Targeted Supervised Contrastive Learning for Long-tailed Recognition (github.com)github.com/LTH14/targeted-supcon 简述 uniformity 是指在理想情况下,监督对比学习应该收敛到一个嵌入,其中不同的类别在超球面上均匀分布。但在长尾学习中,监督对比...
所以在这个系列中,我会系统地解读 Self-Supervised Learning 的经典工作。 Self-Supervised Learning 不仅是在 NLP 领域,在 CV, 语音领域也有很多经典的工作。它可以分成3类:Data Centric, Prediction (也叫 Generative) 和 Contrastive。 其中,Contrastive learning 的范式又叫做 non-parametric instance discrimination,...
. Meanwhile, contrastive loss has been considered instead of the traditional cross-entropy loss in a variety of machine learning applications, showing to be more robust for system stability alternative in self-supervised learning. Following this trend, we hypothesise that using a supervised contrastive...
《Targeted Supervised Contrastive Learning for Long-Tailed Recognition》(CVPR 2022) GitHub: github.com/LTH14/targeted-supcon [fig1]《Aesthetic Text Logo Synthesis via Content-aware Layout Inferring》(CVPR 2022) GitHub: github.com/yizhiwang96/TextLogoLayout...
FTCL:Fine-grained Temporal Contrastive Learning for Weakly-supervised Temporal Action Localization 摘要 我们的目标是弱监督动作定位(WSAL)任务,在模型训练过程中只有视频级的动作标签可用。尽管近年来取得了一些进展,但现有的方法主要遵循于通过优化视频级分类目标来实现定位的方式,这些方法大多忽略了视频之间丰富的时序...
3. Contrastive Regularization for Semi- Supervised Learning In this section, we introduce our contrastive regulariza- tion to improve the SSL performance of the consistency reg- ularization. We first formulate SSL and the consistency reg- ularization, which is the most common a...
3. 对比式学习(Contrastive Learning) 介绍完三个常见的训练策略后,我们至此完成了对图自监督相关的概念,符号等背景知识的介绍,接下来我们将逐个介绍各种方法。由于近一年来Moco [18] 和SimCLR [19] 等算法大火,各种基于互信息最大化的对比学习方法层出不穷,对比式学习的自监督方法最为大家关注和熟悉,我们也将首...
Code:LTH14/targeted-supcon: A PyTorch implementation of the paper Targeted Supervised Contrastive Learning for Long-tailed Recognition (github.com) 图中,随着不平衡比例增加,KCL方法(Ken Chen:论文阅读笔记(1):Exploring Balanced Feature Space For Representation Learning)学习到的不同类比的中心不在均匀分布。