Very, very good research! about the trained model weights

[goodluck110706112]

Thank you very much for your solid research！！
May I ask if the Performance in README refers to the performance of the trained models you provided（load by torch.hub.load method）?

[amaralibey]

Hello @goodluck110706112 ,

Thank you for your kind words and interest in our work!

The performance results mentioned in the README are indeed obtained using the provided weights on Torch Hub. These results are also consistent with those reported in the latest version of our paper on arXiv: https://arxiv.org/abs/2405.07364.

Initially, the paper focused on results using the ResNet-50 backbone only. However, as most recent works used the DINOv2 backbone, we have updated the arXiv version of the paper to include additional results using DINOv2.

Please feel free to reach out if you have more questions or feedback.

Best,

[goodluck110706112]

@amaralibey

ResNet50 + BoQ

vpr_model = torch.hub.load("amaralibey/bag-of-queries", "get_trained_boq", backbone_name="resnet50", output_dim=16384) # dim of embedding is 16384

May I ask if there is a way to use 4096 embeddings instead of 16384?

[amaralibey]

Hello @goodluck110706112

Sorry for the delay, I will soon share the weights of the 4096-d and 2048-d models, for both ResNet and DinoV2.

[goodluck110706112]

@amaralibey Thank you very much for your reply. I expect 4096-d and 2048-d for BoQ~