pretrained senet154 performance seriously drops on Intel(R) Xeon(R) Platinum 8124M CPU @ 3.00GHz

[ghost]

Dear folks,

I know this is not likely your bug, but I think it is good to at least share the result here so other people could avoid wasting time on the same weird issue.

I evaluated senet154 on a small dataset: https://github.com/ajschumacher/imagen

'url': 'http://data.lip6.fr/cadene/pretrainedmodels/senet154-c7b49a05.pth',
'input_space': 'RGB',
'input_size': [3, 224, 224],
'input_range': [0, 1],
'mean': [0.485, 0.456, 0.406],
'std': [0.229, 0.224, 0.225],
'num_classes': 1000

The outputs on most machines are Ok, however the performance on Intel(R) Xeon(R) Platinum 8124M CPU @ 3.00GHz are very poor:

'senet154': '../imagen/imagen/n00007846_147031_person.jpg' is a 'letter opener, paper knife, paperknife'
'senet154': '../imagen/imagen/n00007846_149204_person.jpg' is a 'fig'
'senet154': '../imagen/imagen/n00007846_152343_person.jpg' is a 'letter opener, paper knife, paperknife'
'senet154': '../imagen/imagen/n00007846_160891_person.jpg' is a 'Granny Smith'
'senet154': '../imagen/imagen/n00007846_98724_person.jpg' is a 'scorpion'
'senet154': '../imagen/imagen/n01443537_11099_goldfish.jpg' is a 'television, television system'
'senet154': '../imagen/imagen/n01443537_2625_goldfish.jpg' is a 'Granny Smith'
'senet154': '../imagen/imagen/n01443537_2675_goldfish.jpg' is a 'letter opener, paper knife, paperknife'
'senet154': '../imagen/imagen/n01443537_4691_goldfish.jpg' is a 'letter opener, paper knife, paperknife'
'senet154': '../imagen/imagen/n01443537_5048_goldfish.jpg' is a 'rain barrel'
'senet154': '../imagen/imagen/n01495701_1216_ray.jpg' is a 'rain barrel'
'senet154': '../imagen/imagen/n01495701_1287_ray.jpg' is a 'golf ball'
'senet154': '../imagen/imagen/n01495701_2358_ray.jpg' is a 'chain saw, chainsaw'
'senet154': '../imagen/imagen/n01495701_3614_ray.jpg' is a 'letter opener, paper knife, paperknife'
'senet154': '../imagen/imagen/n01495701_4155_ray.jpg' is a 'letter opener, paper knife, paperknife'
'senet154': '../imagen/imagen/n01503061_10156_bird.jpg' is a 'letter opener, paper knife, paperknife'
'senet154': '../imagen/imagen/n01503061_11000_bird.jpg' is a 'letter opener, paper knife, paperknife'
'senet154': '../imagen/imagen/n01503061_17069_bird.jpg' is a 'letter opener, paper knife, paperknife'

I couldn't reproduce this problem with nasnetalarge,pnasnet5large,resnet152,se_resnet152,se_resnet101 on the same machine. I couldn't reproduce this problem on other machines.

I'll test more and update whatever I found later.

See https://github.com/pytorch/pytorch/issues/18218 for related information.

If you have any idea how to debug I would be happy to try.