search

dangweili / pedestrian-attribute-recognition-pytorch

A simple baseline for pedestrian attribute recognition in surveillance scenarios
334 stars 81 forks source link

about test result #1

Closed hammerwy closed 6 years ago

hammerwy commented 6 years ago

Thanks for your excellent work and code! But i get a issue when testing the model trained by myself using peta dataset, the result is far different from the labels. Do you have the pretrained model work well that i can do some test? Thanks for your reply.

dangweili commented 6 years ago

OK. You can try this one: 链接: https://pan.baidu.com/s/1JfxGDe0Wj2PG-WVHiB394Q 提取码: n4gi The results should be: mA: 0.8423 Acc: 0.7917, Prec: 0.8796, Rec: 0.8559, F1: 0.8676

@Latterminer Due to the bug in transform_peta.py, the results are not exactly as above. I will give a new link to the pretrained model soon.

@trikim Yes. there is a bug in load pretrained model. I will fix it and update the new model.

trikim commented 6 years ago

@dangweili , thanks for providing the pretrained model. I have tested the model by 'sh script/experiment/test.sh' and also get poor performance. I'm doubt the weights are not loaded to the model correctly. what's more, even though the model_weight_file is not exist, the test.sh can run.

dangweili commented 6 years ago

@trikim @Latterminer The new model on peta dataset is here: Links: https://pan.baidu.com/s/17NpH0qeb6Gcap2nFD7XEqQ Code: v9ab

The results should be : mA: 0.8444 Acc: 0.7961, Prec: 0.8800, Rec: 0.8599, F1: 0.8698

Please try it.

trikim commented 6 years ago

@dangweili Thank you very much! The problem is solved. I have trained the model on peta dataset, but get a little poorer performance. The results are: mA: 0.8367 Acc: 0.7873, Prec: 0.8771, Rec: 0.8503, F1: 0.8635 Have you done some other operations(such as data augmentation) except the command "sh ./script/experiment/train.sh" ?

dangweili commented 6 years ago

@trikim I don't use any other operations.

srika91 commented 6 years ago

@dangweili I am not able to download the model from the baidu link.Can u upload it in google drive?

dangweili commented 6 years ago

@srika91 Please try this one: https://drive.google.com/file/d/1_1JeQ3jkuKwgH7NUMai1aMZYh-s0rDie/view?usp=sharing

nakill commented 5 years ago

hi, why Resnet50 (Acc: 0.787 ) < Alexnet (Acc: 82.6 in paper). As far as i know Resnet50 better Alexnet. Thks.

Addie11 commented 5 years ago

Can anyone please share the RAP dataset.

daiwc commented 4 years ago

Can anyone please share the RAP dataset.

if you got rap dataset, please share with me ,thanks a lot!!

BinchaoPeng commented 4 years ago

@dangweili Thank you very much! The problem is solved. I have trained the model on peta dataset, but get a little poorer performance. The results are: mA: 0.8367 Acc: 0.7873, Prec: 0.8771, Rec: 0.8503, F1: 0.8635 Have you done some other operations(such as data augmentation) except the command "sh ./script/experiment/train.sh" ?

do you have the demo about the model，how to use it