mahiro2211/dsh-pytorch

PyTorch implementation of DEEP SUPERVISED HASHING FOR FAST IMAGE RETRIEVAL(CVPR 2016)

DSH_DeepSupervisedHashing

A new version for https://github.com/weixu000/DSH-pytorch

• 2023-6-14说明 ：原有的复现代码并不适合多标签任务分类，需要更改相关损失函数
• 2023-5-21更新 object文件夹可以轻松帮你评估6个指标（mAP,PRcurve,NDCG@1000,TopK-Precision,TopK-Recall,P@H<=2)
• 2023-5-21updated dictory named object will help you easily predict six indexes （mAP,PRcurve,NDCG@1000,TopK-Precision,TopK-Recall,P@H<=2)

Instruction(说明)

• mat文件的保存一定要这样(mat file must be saved like this)
• 可以使用python run.py运行运算指标的程序，但是我最好建议你打开去自己设置它的参数

result_dict = {
        'q_img' : query_img ,
        'r_img' : retrieval_img ,
        'q_l' : query_label ,
        'r_l' : retrieval_label
    }

• PR——cruves是一个计算PR曲线指标的模块 ， save_mat是一个把模型生成的哈希码和标签保存为.mat文件的模块

where you can find the parper

-PyTorch implementation of paper Deep Supervised Hashing for Fast Image Retrieval

It is a brandnew version of DSH_DeepSupervisedHashing

• Which is rewrited by GPT3.5 (I also participated in this job , Because some errors will happen to AI)😆
• 😢 ， but I also modified the parameters in the source code, and a pooling layer in the network ( which can perform a little better than it used to be)

tips you might interested in

• Adding a Savemat function in main.py so that you can use it to Visualize some metrics （ like PrCruve some what .....💐)
• provide a useful method to Visualize Prcruve , U can use Dataspell to do it by using the csv file ##How to Run
• I intended to make it easy for reading. You can easily run it by

python main.py

*commandline uasge have a nice-looking help

pyhon main.py -h

packages

pytorch == 1.12.0 + cu113

torchvision == 0.13.0 + cu113

tqdm

scipy

tornado

matplotlib