why i got torch.Size([701, 512]) torch.Size([51355, 512]) 514 Recall@1:0.00 Recall@5:0.14 Recall@10:0.14 Recall@top1:0.86 AP:0.35

[samcheung1]

I used three_view_long_share_d0.75_256_s1_google model and used test.py, but i got torch.Size([701, 512]) torch.Size([51355, 512]) 514 Recall@1:0.00 Recall@5:0.14 Recall@10:0.14 Recall@top1:0.86 AP:0.35.
That was so bad, so i want to get why?

[layumi]

Hi @samcheung1

The feature dimension seems to be good.
Could you first check the training loss & accuracy? Is it normal?

[samcheung1]

The training loss & accuracy in train.py is normal, but i am not sure the training loss & accuracy in test.py is normal. Can you teach me how?

[layumi]

Hi @samcheung1

1. You may check the loss value and accuracy value during training. The accuracy rate should converge to 90+% after training.
2. Could you print the query and gallery feature in this line?
   https://github.com/layumi/University1652-Baseline/blob/master/test.py#L245-L246

[samcheung1]

Thank you. I got it.

…

---Original--- From: "Zhedong ***@***.***> Date: Mon, Mar 13, 2023 22:40 PM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [layumi/University1652-Baseline] why i got torch.Size([701, 512])torch.Size([51355, 512]) 514 ***@***.***:0.00 ***@***.***:0.14 ***@***.******@***.***:0.86 AP:0.35 (Issue #39) Hi @samcheung1 You may check the loss value and accuracy value during training. The accuracy rate should converge to 90+% after training. Could you print the query and gallery feature in this line? https://github.com/layumi/University1652-Baseline/blob/master/test.py#L245-L246 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>