Question about ResNet pretrained model

bytedance / R2Former

Official repository for R2Former: Unified Retrieval and Reranking Transformer for Place Recognition

Apache License 2.0

83 stars 6 forks source link

Question about ResNet pretrained model #7

Closed jinxxo-j closed 1 year ago

jinxxo-j commented 1 year ago

Thank you for your research! I have a question, when I look at the code, there are ResNet-50 pretrained weights for places dataset and gldv2 dataset, but which ResNet-50 pretrained backbone did you use for ResNet-50 + GeM (Ours w/o reranking) performance in the main paper table 6?

jinxxo-j commented 1 year ago

for training global retrieval model (ResNet-50 + GeM pool), the dimension of feature is 1024?

Jeff-Zilence commented 1 year ago

Can you directly run the code? The implementation of ResNet50+GeM is also included in the code. And the first stage training is based on the Geo-localization benchmark repo with standard setting. Use backbone="resnet50" in the command.