there is a little bug in q2l_infer.py

[Chord-19]

there is a little bug in q2l_infer.py which leads to args.resume will be covered by the "resume " in config.json, and make some people can not repreduct paper's results。 for example：

1. when I train a model ,save its checkpoints in path_1，I use the sentence to begin test: python q2l_infer.py -a 'Q2L-R101-448'
   --config path_1/config.json
   --resume path_1/model_best.pth.tar

2. then the origin args.resume will be replaced by the codes: def parser_args(): if args.config: with open(args.config, 'r') as f: cfg_dict = json.load(f) for k,v in cfg_dict.items(): setattr(args, k, v) return args 3.this may lead model initial without checkpoints.

3. I write the issue because I find some people confuse to get the right results.

[Zhong1015]

Hello, I noticed your comment on CSDN where you pointed out some issues. I see that you've been recently researching this paper, and I'm just starting to delve into this area. I've encountered some difficulties and was hoping to ask you some pressing questions and engage in a discussion. Could we exchange some thoughts?

[Chord-19]

https://github.com/curt-tigges/query2label

------------------ Original ------------------ From: Zhong1015 @.> Date: Tue,Oct 17,2023 0:26 PM To: SlongLiu/query2labels @.> Cc: Chord-19 @.>, Author @.> Subject: Re: [SlongLiu/query2labels] there is a little bug in q2l_infer.py (Issue #47)

Hello, I noticed your comment on CSDN where you pointed out some issues. I see that you've been recently researching this paper, and I'm just starting to delve into this area. I've encountered some difficulties and was hoping to ask you some pressing questions and engage in a discussion. Could we exchange some thoughts?

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

[Zhong1015]

https://github.com/curt-tigges/query2label … ------------------ Original ------------------ From: Zhong1015 @.> Date: Tue,Oct 17,2023 0:26 PM To: SlongLiu/query2labels @.> Cc: Chord-19 @.>, Author @.> Subject: Re: [SlongLiu/query2labels] there is a little bug in q2l_infer.py (Issue #47) Hello, I noticed your comment on CSDN where you pointed out some issues. I see that you've been recently researching this paper, and I'm just starting to delve into this area. I've encountered some difficulties and was hoping to ask you some pressing questions and engage in a discussion. Could we exchange some thoughts? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

Thank you for your reply. I've also seen this blog and its code, but it's a bit cumbersome to use. I would prefer to use the code provided by the original paper's authors. However, in the training part, I can't achieve the expected results; I only get a result of 3.7. I can only obtain normal results in the inference part. I'm not sure where the issue might be.

[Chord-19]

The blog’s codes are easy to use with a little revise in data loader script, I do not use the paper’s origin code.

------------------ Original ------------------ From: Zhong1015 @.> Date: Fri,Oct 27,2023 10:54 AM To: SlongLiu/query2labels @.> Cc: Chord-19 @.>, Author @.> Subject: Re: [SlongLiu/query2labels] there is a little bug in q2l_infer.py (Issue #47)

https://github.com/curt-tigges/query2label … ------------------ Original ------------------ From: Zhong1015 @.> Date: Tue,Oct 17,2023 0:26 PM To: SlongLiu/query2labels @.> Cc: Chord-19 @.>, Author @.> Subject: Re: [SlongLiu/query2labels] there is a little bug in q2l_infer.py (Issue #47) Hello, I noticed your comment on CSDN where you pointed out some issues. I see that you've been recently researching this paper, and I'm just starting to delve into this area. I've encountered some difficulties and was hoping to ask you some pressing questions and engage in a discussion. Could we exchange some thoughts? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

Thank you for your reply. I've also seen this blog and its code, but it's a bit cumbersome to use. I would prefer to use the code provided by the original paper's authors. However, in the training part, I can't achieve the expected results; I only get a result of 3.7. I can only obtain normal results in the inference part. I'm not sure where the issue might be.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

[Zhong1015]

The blog’s codes are easy to use with a little revise in data loader script, I do not use the paper’s origin code. … ------------------ Original ------------------ From: Zhong1015 @.> Date: Fri,Oct 27,2023 10:54 AM To: SlongLiu/query2labels @.> Cc: Chord-19 @.>, Author @.> Subject: Re: [SlongLiu/query2labels] there is a little bug in q2l_infer.py (Issue #47) https://github.com/curt-tigges/query2label … ------------------ Original ------------------ From: Zhong1015 @.> Date: Tue,Oct 17,2023 0:26 PM To: SlongLiu/query2labels @.> Cc: Chord-19 @.>, Author @.> Subject: Re: [SlongLiu/query2labels] there is a little bug in q2l_infer.py (Issue #47) Hello, I noticed your comment on CSDN where you pointed out some issues. I see that you've been recently researching this paper, and I'm just starting to delve into this area. I've encountered some difficulties and was hoping to ask you some pressing questions and engage in a discussion. Could we exchange some thoughts? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.> Thank you for your reply. I've also seen this blog and its code, but it's a bit cumbersome to use. I would prefer to use the code provided by the original paper's authors. However, in the training part, I can't achieve the expected results; I only get a result of 3.7. I can only obtain normal results in the inference part. I'm not sure where the issue might be. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.>

The evaluation metrics provided in that blog post, after my verification, turned out to be incorrect. There is a decrease of 3-4 percentage points compared to the actual mAP (mean Average Precision), which is very troubling for me.

[ly-pear]

Hello, may I ask how to solve this bug