keras models summary

This repository contains python jupyter notebooks of keras models with their summaries. In order for someone to view a model summary, has to load the model and run model summary. As simple as it is, it can be time-consuming when you are looking for the right model and have to load one model after another.

Summary Informations

In Keras model summary we can find informations about:

• number pamameters
• default input size
• layer name and shape

Keras pretrained models origin

Keras Models are loaded either from official keras library or from classification_models library. Keras library's collection of pre-trained models doesn' t contain some models that can be useful at this point. classification_models library contains some models that keras doesn't but also contains weights from other sources too (ex. resnet50 weights from mxnet).

Num Features and Parameteres after applying a pooling layer

From keras package (keras.applications)

Model	Features	Size	Stem	input size (default)
DenseNet121	1024	8.1M	7.0M	224
DenseNet169	1664	14.3M	12.6M	224
DenseNet201	1920	20.2M	18.3M	224
InceptionResNetV2	1536	55.9M	54.3M	299
InceptionV3	2048	23.9M	21.8M	299
MobileNet(alpha=1.0)	1024	4.3M	3.2M	224
MobileNetV2(alpha=1.0)	1024	3.5M	2.3M	224
NASNetLarge	4032	93.5M	84.9M	331
NASNetMobile	1056	7.7M	4.3M	224
Resnet50	2048	25.6M	23.6M	224
VGG16	512	138.4M	14.7M	224
VGG19	512	143.7M	20.0M	224
Xception	2048	22.9M	20.9M	299

From image-classifiers package (classification_models)

Model	Features	Size	Stem
Resnet18	512	11.7M	11.2M
Resnet34	512	21.8M	21.3M
Resnet101	2048	44.6M	42.6M
Resnet152	2048	60.3M	58.3M
Resnext50	2048	25.1M	23.0M
Resnext101	2048	44.3M	42.3M
Senet154	2048	115.3M	113.3M
SeResnet18	512	11.8M	11.3M
SeResnet34	512	22.0M	21.5M
SeResnet50	2048	28.1M	26.1M
SeResnet101	2048	49.4M	47.4M
SeResnet152	2048	67.0M	64.9M
SeResnext50	2048	27.6M	25.6M
SeResnext101	2048	49.1M	47.0M