search

SikaStar / IDM

MIT License
100 stars 19 forks source link

issue about reproduced results #3

Closed zaiquanyang closed 3 years ago

zaiquanyang commented 3 years ago

When I use the given example as follow

# IBN-ResNet-50 + IDM
CUDA_VISIBLE_DEVICES=0,1,2,3 \
sh scripts/run_idm_xbm.sh market1501 dukemtmc resnet_ibn50a_idm 0 0.7 0.1 1.0

, I can not reproduce the reported mAP by a margin of 9 \% .
Looking forward to your reply.

SikaStar commented 3 years ago

When I use the given example as follow

# IBN-ResNet-50 + IDM
CUDA_VISIBLE_DEVICES=0,1,2,3 \
sh scripts/run_idm_xbm.sh market1501 dukemtmc resnet_ibn50a_idm 0 0.7 0.1 1.0

, I can not reproduce the reported mAP by a margin of 9 % . Looking forward to your reply.

Hi, do you mean that you got mAP=62.1%? If so, maybe you have not downloaded the Imagenet-pretrained model from this link and save it under the path of logs/pretrained/

zaiquanyang commented 3 years ago

When I use the given example as follow

# IBN-ResNet-50 + IDM
CUDA_VISIBLE_DEVICES=0,1,2,3 \
sh scripts/run_idm_xbm.sh market1501 dukemtmc resnet_ibn50a_idm 0 0.7 0.1 1.0

, I can not reproduce the reported mAP by a margin of 9 % . Looking forward to your reply.

Hi, do you mean that you got mAP=62.1%? If so, maybe you have not downloaded the Imagenet-pretrained model from this link and save it under the path of logs/pretrained/

I indeed followed your instructions to download the network and put it in the specified location (in my machine) as follow:

/data/code/DA_ReID/IDM/logs/pretrained/resnet50_ibn_a.pth.tar

and I train the model about 40 epochs.

SikaStar commented 3 years ago 
/data/code/DA_ReID/IDM/logs/pretrained/resnet50_ibn_a.pth.tar

What are the results you have got?

zaiquanyang commented 3 years ago 
/data/code/DA_ReID/IDM/logs/pretrained/resnet50_ibn_a.pth.tar

What are the results you have got?

Mean AP: 55.5% CMC Scores: top-1 74.7% top-5 84.5% top-10 87.1%

SikaStar commented 3 years ago 
/data/code/DA_ReID/IDM/logs/pretrained/resnet50_ibn_a.pth.tar

What are the results you have got?

Mean AP: 55.5% CMC Scores: top-1 74.7% top-5 84.5% top-10 87.1%

  • Finished epoch 44 model mAP: 55.5% best: 61.0%

It is strange! Have you trained with 4 GPUs?

zaiquanyang commented 3 years ago 
/data/code/DA_ReID/IDM/logs/pretrained/resnet50_ibn_a.pth.tar

What are the results you have got?

Mean AP: 55.5% CMC Scores: top-1 74.7% top-5 84.5% top-10 87.1%

  • Finished epoch 44 model mAP: 55.5% best: 61.0%

It is strange! Have you trained with 4 GPUs?

I just run it on a v100GPU

SikaStar commented 3 years ago 
/data/code/DA_ReID/IDM/logs/pretrained/resnet50_ibn_a.pth.tar

What are the results you have got?

Mean AP: 55.5% CMC Scores: top-1 74.7% top-5 84.5% top-10 87.1%

  • Finished epoch 44 model mAP: 55.5% best: 61.0%

It is strange! Have you trained with 4 GPUs?

I just run it on a v100GPU

Hi, I have not conducted these experiments on a single GPU. All my experiments are trained on 4 GPUs. As far as I know, the number of GPUs can affect the results significantly. You can try it on 4 GPUs and you will get similar results that I reported in my paper.

zaiquanyang commented 3 years ago 
/data/code/DA_ReID/IDM/logs/pretrained/resnet50_ibn_a.pth.tar

What are the results you have got?

Mean AP: 55.5% CMC Scores: top-1 74.7% top-5 84.5% top-10 87.1%

  • Finished epoch 44 model mAP: 55.5% best: 61.0%

It is strange! Have you trained with 4 GPUs?

I just run it on a v100GPU

Hi, I have not conducted these experiments on a single GPU. All my experiments are trained on 4 GPUs. As far as I know, the number of GPUs can affect the results significantly. You can try it on 4 GPUs and you will get similar results that I reported in my paper.

ok, thanks very much for your reply!