2026-10-15T00:36:53Z https://www.scidb.cn/oai

10.57760/sciencedb.43792 2026-07-28T17:00:36Z

2026-07-28 SoftRankCLIP: Soft Top-K Aggregation for Efficient Image-Text Retrieval doi:10.57760/sciencedb.43792 en CLIP-style dual encoders support efficient image-text retrieval through embedding-only inference. However, common diagonal contrastive training may underuse the positive relationships available in captioned datasets, where each image is associated with multiple valid descriptions. We present SoftRankCLIP, a lightweight fine-tuning formulation that combines a multi-positive contrastive objective with a smooth negative-aware separation loss and a temperature-weighted top-$K$ positive score. All additional objectives are used only during training, leaving the encoder architecture and retrieval procedure unchanged at inference. On Flickr30k and MS-COCO, the complete objective obtains higher retrieval scores than the corresponding pretrained references and the alternative fine-tuning objectives evaluated in our study. Ablation, backbone-transfer, and efficiency results further indicate that the training formulation is portable across the tested encoders and introduces moderate training overhead. These results provide empirical evidence that ranking-aware multi-positive objectives can improve CLIP-style retrieval without modifying embedding-only inference. segmentation; and captioning dataset.; large-scale object detection; CLIP; image-text retrieval; multi-positive contrastive 李斌斌 郝庭钰 亓子森 PUBLIC https://api.github.com/licenses/cc0-1.0 dataset Science Data Bank