What are the meanings of all the model names in the timm library?

Deep Learning
1

If I want to use a particular model from the timm library for a problem, how do I know what all the models mean?

For example, here are the variants for the MobileNet model.

If I run-

import timm
from pprint import pprint
model_names = timm.list_models(pretrained=True)
pprint(model_names)

I get a long list, and the models related to the MobileNet architecture are-

‘mobilenetv2_100’,
‘mobilenetv2_110d’,
‘mobilenetv2_120d’,
‘mobilenetv2_140’,
‘mobilenetv3_large_100’,
‘mobilenetv3_large_100_miil’,
‘mobilenetv3_large_100_miil_in21k’,
‘mobilenetv3_rw’,

How do I know what those mean?

What does rw mean? What do those ds at the end of the model names mean? And what exactly does large mean? I guess the numbers with miil could be the number of parameters. What is _in21k at the end of one model name?

How do I know these?

I am very new to timm, and please suggest helpful resources at will.

It would be also very helpful if someone gives me a list of all the **kwargs available to the create_model() method.

Thanks.

2

in2k models were trained on ImageNet21k (imagenet with 21k classes instead of 1k). Besides that i‘m also a bit confused about the model names in timm :blush:

1 Like
3

I am trying to figure out as well. I think rw means Ross Wightman’s version. For example, found this on the timm’s GH readme Rename my prev V2 guess efficientnet_v2s -> efficientnetv2_rw_s. Can be wrong but some resources that I found helpful

https://rwightman.github.io/pytorch-image-models/models/