April 11, 2023
We introduce Patch Aligned Contrastive Learning (PACL), a modified compatibility function for CLIP’s con- trastive loss, intending to train an alignment between the patch tokens of the vision encoder and the CLS token of the text encoder. With such an alignment, a model can identify regions of an image corresponding to a given text input, and therefore transfer seamlessly to the task of open vocabulary semantic segmentation without requiring any segmentation annotations during training. Using pre-trained CLIP en- coders with PACL, we are able to set the state-of-the-art on the task of open vocabulary zero-shot segmentation on 4 different segmentation benchmarks: Pascal VOC, Pascal Context, COCO Stuff and ADE20K. Furthermore, we show that PACL is also applicable to image-level predictions and when used with a CLIP backbone, provides a general im- provement in zero-shot classification accuracy compared to CLIP, across a suite of 12 image classification datasets.
Written by
Jishnu Mukhoti
Tsung-Yu Lin
Omid Poursaeed
Rui (VisionX) Wang
Ashish Shah
Philip Torr
Publisher
CVPR
Research Topics
September 30, 2023
Pierre Fernandez, Guillaume Couairon, Hervé Jegou, Matthijs Douze, Teddy Furon
September 30, 2023
September 29, 2023
Yiming Li, Qi Fang, Jiamu Bai, Siheng Chen, Felix Xu, Chen Feng
September 29, 2023
September 27, 2023
Xiaoliang Dai, Ji Hou, Kevin Chih-Yao Ma, Sam Tsai, Jialiang Wang, Rui Wang, Peizhao Zhang, Simon Vandenhende, Xiaofang Wang, Abhimanyu Dubey, Matthew Yu, Abhishek Kadian, Filip Radenovic, Dhruv Mahajan, Kunpeng Li, Yue (R) Zhao, Vladan Petrovic, Mitesh Kumar Singh, Simran Motwani, Yiwen Song, Yi Wen, Roshan Sumbaly, Vignesh Ramanathan, Zijian He, Peter Vajda, Devi Parikh
September 27, 2023
September 22, 2023
Shuangzhi Li, Zhijie Wang, Felix Xu, Qing Guo, Xingyu Li, Lei Ma
September 22, 2023
Who We Are
Our Actions
Newsletter