exponent of param_count in csv results file

[ardeal]

Hi,

I checked the results.csv file. For resnet18, the param_count is:

resnet18	69.748	30.252	89.084	10.916	304.76	224	0.875	bilinear

my question is:
what is exponent of the number 304.76?
as we know, the parameter count of resnet18 is around 31Million which is 31x1e6

[rwightman]

@ardeal where is there a row with 304.76 as the resnet18 param count? it's million, 1e6 and resnet18 is 10.69

https://github.com/rwightman/pytorch-image-models/blob/master/results/results-imagenet.csv#L650

[ardeal]

@rwightman ,

there are a few questions:

1. the parameters count of resnet18 should be approximate 3million. according to your explanation, it is 10.69X1e6.
2. for your data, 10.69 is not param_count, but 10.69 is top5_err.
3. yesterday, I git clone the repo, and I capture the data in csv file as following picture. The param_count is 304.76.

[rwightman]

@ardeal sorry typo in my resppnse, 11.69. You've messed up sorting your csv, you need to extend sort to all columns, you only sorted param count...

[ardeal]

@rwightman ,
No worries!
Let's ignore the sorting thing.
for the data 11.69, it means that the param_count of resnet18 is 11.69x1e6 which is not approximate identical 3Million.

[ardeal]

@rwightman
Hi,
Do you have any ideas about my question above?