details about the base model and the fine-tuned about fewshot_detection

[cgg99]

@futureisatyourhand Hi~How did you evaluate the base model？Did you execute python valid_ensemble.py cfg/metayolo_novel0.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg path/toweightfile ？

[futureisatyourhand]

@futureisatyourhand Hi~How did you evaluate the base model？Did you execute python valid_ensemble.py cfg/metayolo_novel0.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg path/toweightfile ？

Hello, I have solved the problem, but the AP is much higher than the author's, the randomness is very strong.

[MonsterLWR]

i've met similar problem, could you please tell us how you solve it? @futureisatyourhand
thanks!

[futureisatyourhand]

i've met similar problem, could you please tell us how you solve it? @futureisatyourhand
thanks!
First of all, the author's code is very random, the AP after each training is different, you'd better set up a random seed.
Second, the rest of the git code cannot be changed, you can try to retrain the git author's code, you may find that the results are better than the author's code, the author's code is very metaphysical, I just re-git the author's code and training results are better than the author.
For details, you can look at the results that I reproduced and optimized the algorithm, the linke is https://arxiv.org/pdf/2007.06837.pdf.

[MonsterLWR]

Thanks for your reply! I still have some questions:
how many gpus did you use? when using two gpus, should i change the learning rate or max batches? @futureisatyourhand

[GDDKRJ]

Hi~How did you perform the code when evaluating the model？
“python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg path/toweightfile
python scripts/voc_eval.py results/path/to/comp4_det_test_”
I don’t know the “path/toweightfile” and “results/path/to/comp4_det_test_”

[futureisatyourhand]

Hi~How did you perform the code when evaluating the model？
“python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg path/toweightfile
python scripts/voc_eval.py results/path/to/comp4_det_test_”
I don’t know the “path/toweightfile” and “results/path/to/comp4_det_test_”

The weight file is your best weights saved, and the "results/path/to/comp4_det_test_" is your results containing boxes and label. when the first instruction executed successfully, and the the "results/path/to/comp4_det_test_" is generated. You should scan source code.

[GDDKRJ]

Is the first instruction written like this? python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg backup/metayolo/000010.weights I want to know the best weights is darknet19_448.cfg or backup/metayolo/****.weights? Because when I use darknet19_448.cfg the code can't run

…

------------------ 原始邮件 ------------------ 发件人: "bingykang/Fewshot_Detection" ***@***.***>; 发送时间: 2021年9月23日(星期四) 上午9:19 ***@***.***>; ***@***.******@***.***>; 主题: Re: [bingykang/Fewshot_Detection] details about the base model and the fine-tuned (#18) Hi~How did you perform the code when evaluating the model？ “python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg path/toweightfile python scripts/voc_eval.py results/path/to/comp4_det_test_” I don’t know the “path/toweightfile” and “results/path/to/comp4_det_test_” The weight file is your best weights saved, and the "results/path/to/comp4_det_test_" is your results containing boxes and label. when the first instruction executed successfully, and the the "results/path/to/comp4_det_test_" is generated. You should scan source code. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

[futureisatyourhand]

Is the first instruction written like this? python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg backup/metayolo/000010.weights I want to know the best weights is darknet19_448.cfg or backup/metayolo/.weights? Because when I use darknet19_448.cfg the code can't run
…
------------------ 原始邮件 ------------------ 发件人: "bingykang/Fewshot_Detection" @.>; 发送时间: 2021年9月23日(星期四) 上午9:19 @.>; @.@.**>; 主题: Re: [bingykang/Fewshot_Detection] details about the base model and the fine-tuned (#18) Hi~How did you perform the code when evaluating the model？ “python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg path/toweightfile python scripts/voc_eval.py results/path/to/comp4_det_test_” I don’t know the “path/toweightfile” and “results/path/to/comp4_det_test_” The weight file is your best weights saved, and the "results/path/to/comp4_det_test_" is your results containing boxes and label. when the first instruction executed successfully, and the the "results/path/to/comp4_det_test_" is generated. You should scan source code. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

the

Is the first instruction written like this? python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg backup/metayolo/000010.weights I want to know the best weights is darknet19_448.cfg or backup/metayolo/.weights? Because when I use darknet19_448.cfg the code can't run
…
------------------ 原始邮件 ------------------ 发件人: "bingykang/Fewshot_Detection" @.>; 发送时间: 2021年9月23日(星期四) 上午9:19 @.>; @.@.**>; 主题: Re: [bingykang/Fewshot_Detection] details about the base model and the fine-tuned (#18) Hi~How did you perform the code when evaluating the model？ “python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg path/toweightfile python scripts/voc_eval.py results/path/to/comp4_det_test_” I don’t know the “path/toweightfile” and “results/path/to/comp4_det_test_” The weight file is your best weights saved, and the "results/path/to/comp4_det_test_" is your results containing boxes and label. when the first instruction executed successfully, and the the "results/path/to/comp4_det_test_" is generated. You should scan source code. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

the instruction is to test, so it is no relation to darknet19_448.cfg, the darknet19_448.weights is pretrained weights to get backup/metayolo/000010.weights!!!

[GDDKRJ]

I want to know why the ***.txt is empty?It has no information.

…

------------------ 原始邮件 ------------------ 发件人: "bingykang/Fewshot_Detection" ***@***.***>; 发送时间: 2021年9月23日(星期四) 中午11:40 ***@***.***>; ***@***.******@***.***>; 主题: Re: [bingykang/Fewshot_Detection] details about the base model and the fine-tuned (#18) Is the first instruction written like this? python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg backup/metayolo/000010.weights I want to know the best weights is darknet19_448.cfg or backup/metayolo/.weights? Because when I use darknet19_448.cfg the code can't run … ------------------ 原始邮件 ------------------ 发件人: "bingykang/Fewshot_Detection" @.>; 发送时间: 2021年9月23日(星期四) 上午9:19 @.>; @.@.**>; 主题: Re: [bingykang/Fewshot_Detection] details about the base model and the fine-tuned (#18) Hi~How did you perform the code when evaluating the model？ “python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg path/toweightfile python scripts/voc_eval.py results/path/to/comp4_det_test_” I don’t know the “path/toweightfile” and “results/path/to/comp4_det_test_” The weight file is your best weights saved, and the "results/path/to/comp4_det_test_" is your results containing boxes and label. when the first instruction executed successfully, and the the "results/path/to/comp4_det_test_" is generated. You should scan source code. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android. the Is the first instruction written like this? python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg backup/metayolo/000010.weights I want to know the best weights is darknet19_448.cfg or backup/metayolo/.weights? Because when I use darknet19_448.cfg the code can't run … ------------------ 原始邮件 ------------------ 发件人: "bingykang/Fewshot_Detection" @.>; 发送时间: 2021年9月23日(星期四) 上午9:19 @.>; @.@.**>; 主题: Re: [bingykang/Fewshot_Detection] details about the base model and the fine-tuned (#18) Hi~How did you perform the code when evaluating the model？ “python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg path/toweightfile python scripts/voc_eval.py results/path/to/comp4_det_test_” I don’t know the “path/toweightfile” and “results/path/to/comp4_det_test_” The weight file is your best weights saved, and the "results/path/to/comp4_det_test_" is your results containing boxes and label. when the first instruction executed successfully, and the the "results/path/to/comp4_det_test_" is generated. You should scan source code. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android. the instruction is to test, so it is no relation to darknet19_448.cfg, the darknet19_448.weights is pretrained weights to get backup/metayolo/000010.weights!!! — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

[futureisatyourhand]

I want to know why the .txt is empty?It has no information.
…
------------------ 原始邮件 ------------------ 发件人: "bingykang/Fewshot_Detection" @.>; 发送时间: 2021年9月23日(星期四) 中午11:40 @.>; @.@.>; 主题: Re: [bingykang/Fewshot_Detection] details about the base model and the fine-tuned (#18) Is the first instruction written like this? python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg backup/metayolo/000010.weights I want to know the best weights is darknet19_448.cfg or backup/metayolo/.weights? Because when I use darknet19_448.cfg the code can't run … ------------------ 原始邮件 ------------------ 发件人: "bingykang/Fewshot_Detection" @.>; 发送时间: 2021年9月23日(星期四) 上午9:19 @.>; @.@.>; 主题: Re: [bingykang/Fewshot_Detection] details about the base model and the fine-tuned (#18) Hi~How did you perform the code when evaluating the model？ “python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg path/toweightfile python scripts/voc_eval.py results/path/to/comp4_det_test_” I don’t know the “path/toweightfile” and “results/path/to/comp4_det_test_” The weight file is your best weights saved, and the "results/path/to/comp4_det_test_" is your results containing boxes and label. when the first instruction executed successfully, and the the "results/path/to/comp4_det_test_" is generated. You should scan source code. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android. the Is the first instruction written like this? python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg backup/metayolo/000010.weights I want to know the best weights is darknet19_448.cfg or backup/metayolo/.weights? Because when I use darknet19_448.cfg the code can't run … ------------------ 原始邮件 ------------------ 发件人: "bingykang/Fewshot_Detection" @.>; 发送时间: 2021年9月23日(星期四) 上午9:19 @.>; @.@.*>; 主题: Re: [bingykang/Fewshot_Detection] details about the base model and the fine-tuned (#18) Hi~How did you perform the code when evaluating the model？ “python valid_ensemble.py cfg/metayolo.data cfg/darknet_dynamic.cfg cfg/reweighting_net.cfg path/toweightfile python scripts/voc_eval.py results/path/to/comp4_det_test_” I don’t know the “path/toweightfile” and “results/path/to/comp4_det_test_” The weight file is your best weights saved, and the "results/path/to/comp4_det_test_" is your results containing boxes and label. when the first instruction executed successfully, and the the "results/path/to/comp4_det_test_" is generated. You should scan source code. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android. the instruction is to test, so it is no relation to darknet19_448.cfg, the darknet19_448.weights is pretrained weights to get backup/metayolo/000010.weights!!! — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

your result is none! You must look code.