Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

MaskCLIP: Masked Self-Distillation Advances Contrastive Language-Image Pretraining

About

This paper presents a simple yet effective framework MaskCLIP, which incorporates a newly proposed masked self-distillation into contrastive language-image pretraining. The core idea of masked self-distillation is to distill representation from a full image to the representation predicted from a masked image. Such incorporation enjoys two vital benefits. First, masked self-distillation targets local patch representation learning, which is complementary to vision-language contrastive focusing on text-related representation. Second, masked self-distillation is also consistent with vision-language contrastive from the perspective of training objective as both utilize the visual encoder for feature aligning, and thus is able to learn local semantics getting indirect supervision from the language. We provide specially designed experiments with a comprehensive analysis to validate the two benefits. Symmetrically, we also introduce the local semantic supervision into the text branch, which further improves the pretraining performance. With extensive experiments, we show that MaskCLIP, when applied to various challenging downstream tasks, achieves superior results in linear probing, finetuning, and zero-shot performance with the guidance of the language encoder. Code will be release at \url{https://github.com/LightDXY/MaskCLIP}.

Xiaoyi Dong, Jianmin Bao, Yinglin Zheng, Ting Zhang, Dongdong Chen, Hao Yang, Ming Zeng, Weiming Zhang, Lu Yuan, Dong Chen, Fang Wen, Nenghai Yu• 2022

Related benchmarks

TaskDatasetResultRank
Semantic segmentationADE20K (val)
mIoU50.5
2888
Instance SegmentationCOCO 2017 (val)
APm0.409
1201
Semantic segmentationADE20K
mIoU11.9
1024
Semantic segmentationCityscapes
mIoU24.9
658
Image ClassificationImageNet-1k (val)
Top-1 Accuracy83.6
543
Text-to-Image RetrievalFlickr30K
R@145.6
531
Semantic segmentationCOCO Stuff
mIoU16.7
379
Semantic segmentationPASCAL Context (val)
mIoU16.8
360
Object DetectionMS-COCO 2017 (val)--
237
Semantic segmentationPascal Context
mIoU17.2
217
Showing 10 of 45 rows

Other info

Code

Follow for update