Benyi Hu

Results 8 comments of Benyi Hu

About different size feature (extract_features) with ["all"]

1) "all" is used to control the output features in extracting, so feature dimension is not in this situation; 2) In indexing phase, you should assign the specific feature names,...

Preparing the custom dataset

There are two way for you: 1) change the data structure to the form we used now 2) add your own method to read your dataset in the pyretri/extract/utils/make_data_json.py

How to use GAP and GMP simultaneously?

What is the 'output' in your feature names? Pyretri will automatically concatenate multiple features in the L dimension.

aqe(query_feat: torch.tensor, gallery_feat: torch.tensor, qe_times: int = 1, qe_k: int = 10, alpha: float = 3.0):

您好，能告诉我具体的代码位置吗？

Suggestion for improving pairwise distance calculation

Thanks for you suggestions! I will fix it in the next vesion.

pre-processing&data augmentation

To get better feature representation :)

How to get the same performance as SCDA paper in CUB200-2011 dataset?

For SCDA, it calculate the top-5 mAP, we calculate mAP on all the samples, so we can not get the same performance.

Possible bug in k reciprocal computation

Thanks for you suggestions! I will fix it in the next version.