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

Aligning Information Capacity Between Vision and Language via Dense-to-Sparse Feature Distillation for Image-Text Matching

About

Enabling Visual Semantic Models to effectively handle multi-view description matching has been a longstanding challenge. Existing methods typically learn a set of embeddings to find the optimal match for each view's text and compute similarity. However, the visual and text embeddings learned through these approaches have limited information capacity and are prone to interference from locally similar negative samples. To address this issue, we argue that the information capacity of embeddings is crucial and propose Dense-to-Sparse Feature Distilled Visual Semantic Embedding (D2S-VSE), which enhances the information capacity of sparse text by leveraging dense text distillation. Specifically, D2S-VSE is a two-stage framework. In the pre-training stage, we align images with dense text to enhance the information capacity of visual semantic embeddings. In the fine-tuning stage, we optimize two tasks simultaneously, distilling dense text embeddings to sparse text embeddings while aligning images and sparse texts, enhancing the information capacity of sparse text embeddings. Our proposed D2S-VSE model is extensively evaluated on the large-scale MS-COCO and Flickr30K datasets, demonstrating its superiority over recent state-of-the-art methods.

Yang Liu, Wentao Feng, Zhuoyao Liu, Shudong Huang, Jiancheng Lv• 2025

Related benchmarks

TaskDatasetResultRank
Image-to-Text RetrievalFlickr30K 1K (test)
R@183.1
491
Text-to-Image RetrievalFlickr30K 1K (test)
R@168.5
432
Image-to-Text RetrievalMS-COCO 5K (test)
R@160.1
320
Text-to-Image RetrievalMS-COCO 5K (test)
R@146.3
244
Image-to-Text RetrievalCamoIT 3K (test)
R@113.3
15
Text-to-Image RetrievalCamoIT 3K (test)
R@112.7
15
Multimodal RetrievalMovies MAG benchmark
Recall@154.5
13
Multimodal RetrievalRedditS MAG benchmark
R@19.4
13
Multimodal RetrievalEle-fashion MAG benchmark
R@115.6
13
Multimodal RetrievalMAG Toys benchmark
R@150
13
Showing 10 of 15 rows

Other info

Follow for update