search

MCC-WH / SSP

Official implementation of paper "Structure Similarity Preservation Learning for Asymmetric Image Retrieval"
8 stars 1 forks source link

How to get R1M_R101_GeM.pkl and R1M_R101_DELG.pkl ? #1

Closed SeunghanYu closed 8 months ago

SeunghanYu commented 8 months ago

Hi, @MCC-WH, I'm encountering an issue in the SSP code related to the generation of anchor points. For generating anchor points, this code requires files named R1M_R101_GeM.pkl and R1M_R101_DELG.pkl, which I can't find. Could you provide the separate code to generate these files? Alternatively, is it possible to obtain the R1M_R101_GeM.pkl or R1M_R101_DELG.pkl files directly?

Looking forward to your reply. Thank you in advance!

MCC-WH commented 8 months ago

Thanks for your interesting. You can fine the code and model at Filip Radenovic's great repository on image retrieval. For 'R1M_R101_GeM.pkl', the model is released at R101-DELG.

SeunghanYu commented 8 months ago

@MCC-WH

Thank you for your quick response. I have another question regarding the process of creating the R1M_R101_GeM.pkl file.

The R1M dataset consists of 1,001,001 images. Is it correct to extract the image features from all 1,001,001 images, save them into the R1M_R101_GeM.pkl file, and then use this file to generate Anchor Points?"

MCC-WH commented 8 months ago

Yes, and you can also sample some images to generate anchor points, which saves time and computational overhead.

SeunghanYu commented 8 months ago

Thank you! I made a code to extract R1M features and successfully get R1M_R101_GeM.pkl and R1M_R101_DELG.pkl.