Unsupervised Multi-semantic Similarity Reconstruction Contrastive Hashing for Multi-label Image Retrieval

This is the implementation of paper: Unsupervised Multi-semantic Similarity Reconstruction Contrastive Hashing for Multi-label Image Retrieval.(DOI: 10.1016/j.eswa.2025.130414)

Datasets

Experiments on 3 image datasets: FLICKR25K, COCO2014, NUSWIDE.

Dependencies

• Python 3.8
• Pytorch 1.10.1
• torchvision 0.11.2
• CUDA 11.1

Training Example

python train.py with umrch flickr hash_bit=16

Model Weights

Download: Link

Test Example

python metric.py --dataset flickr --hash_bit 16 --method umrch --iscode --backbone_frozen --ckpt checkpoints/umrch_logs/best_umrch_100_flickr_clip_True_16_32/*_best.pth