DEnse Contrastive

Dense contrastive learning (DCL) aims to learn rich, spatially-aware representations from unlabeled data by contrasting features at the pixel or patch level, improving upon methods that only consider global image representations. Current research focuses on refining the methods for creating positive and negative pairs of features, exploring different matching strategies (e.g., precise location matching, patch-level contrasting without explicit correspondence), and analyzing the relationship between representation properties (alignment and uniformity) and downstream task performance. These advancements significantly enhance performance on dense prediction tasks like segmentation and object detection in various domains, including image classification and medical image analysis, demonstrating the value of DCL for self-supervised learning.

Papers

June 23, 2023

Patch-Level Contrasting without Patch Correspondence for Accurate and Dense Contrastive Representation Learning
Shaofeng Zhang, Feng Zhu, Rui Zhao, Junchi Yan
Supervised ImageNet Contrastive Method DEnse Contrastive Spatial Aware Dense Visual Representation

December 23, 2022

Precise Location Matching Improves Dense Contrastive Learning in Digital Pathology
Jingwei Zhang, Saarthak Kapse, Ke Ma, Prateek Prasanna, Maria Vakalopoulou, Joel Saltz, Dimitris Samaras
Contrastive Learning Digital Pathology Dense Prediction Precise Localization Dense Feature Dense Matching DEnse Contrastive

October 26, 2022

IDEAL: Improved DEnse locAL Contrastive Learning for Semi-Supervised Medical Image Segmentation
Hritam Basak, Soumitri Chattopadhyay, Rohit Kundu, Sayan Nag, Rammohan Mallipeddi
Metric Learning Pixel Level Semi Supervised Medical Image Segmentation Contrastive Self Supervised Learning Cardiac MRI Segmentation DEnse Contrastive

October 17, 2022

Correlation between Alignment-Uniformity and Performance of Dense Contrastive Representations
Jong Hak Moon, Wonjae Kim, Edward Choi
System Performance Total Correlation Uniformity Metric Contrastive Representation Instance Level Contrastive DEnse Contrastive

October 11, 2022

Improving Dense Contrastive Learning with Dense Negative Pairs
Berk Iskender, Zhenlin Xu, Simon Kornblith, En-Hung Chu, Maryam Khademi
Contrastive Representation Learning Better Representation Negative Pair DEnse Contrastive

July 8, 2022

Pixel-level Correspondence for Self-Supervised Learning from Video
Yash Sharma, Yi Zhu, Chris Russell, Thomas Brox
Self Supervised Learning Optical Flow Source Video Supervised Baseline Pixel Correspondence DEnse Contrastive

DEnse Contrastive

Papers

Patch-Level Contrasting without Patch Correspondence for Accurate and Dense Contrastive Representation Learning

Precise Location Matching Improves Dense Contrastive Learning in Digital Pathology

IDEAL: Improved DEnse locAL Contrastive Learning for Semi-Supervised Medical Image Segmentation

Correlation between Alignment-Uniformity and Performance of Dense Contrastive Representations

Improving Dense Contrastive Learning with Dense Negative Pairs

Pixel-level Correspondence for Self-Supervised Learning from Video