zbr17 / drml

[ICCV 2021] Official PyTorch implementation for Deep Relational Metric Learning.

Hi there 👋

drml's Issues

TypeError: forward() got an unexpected keyword argument 'features'

while running
CUDA_VISIBLE_DEVICES=1 python examples/demo.py --data_path ./ --save_path ./output --device 0 --batch_size 180 --test_batch_size 180 --setting proxy_anchor_drml --embeddings_dim 512 --features_dim 1024 --branch_num 4 --proxyanchor_margin 0.2 --proxyanchor_alpha 64 --num_classes 100 --wd 0.0001 --gamma 0.5 --step 10 --lr_trunk 0.00005 --lr_embedder 0.001 --lr_collector 0.01 --weight_recon_loss 1 --weight_repre_loss 10 --dataset cub200 --delete_old --warm_up 1 --warm_up_list embedder ensemble repre

学长，thudml这个包下载不了啊

[Fail Examples] Can I see the fail examples?

Hi! Thanks for your interesting works.

I am a beginner in the field of Deep Metal Learning.

From the code, Recall@1 is calculated. I would like to ask how to see images fail to be recalled ( fail examples in Recall@1). That is, when the DRML model searches for an image, I want to see images that have not been successfully searched. Which file should I write this code from?

Once again, I'm a very beginner. The scale of this code feels large, and it is difficult to grasp the structure. If you answer, it will be of great help. Thank you.

How to use these ckpt files? How to generate the embedding features of an image?

Hi,
I am doing research on metric learning recently! When retraining the model, I encountered a problem. I got a lot of ckpt files.
I'm confused about which file to use. And how to generate the embedding features of an image? Will test scripts be open sourced
recently?
Thank you very much!