Available architectures

Here is the complete list of all the neural network architectures available in Studio. When available, links to the research papers are provided.

Classification and tagging architectures

Backbone

Trained on

Input size

Batch size

Research paper

Inception-Resnet v2

ImageNet

299x299

32

arXiv

ResNet-50 v2

ImageNet

224x224

32

arXiv

ResNet-101 v2

ImageNet

224x224

32

arXiv

ResNet-152 v2

ImageNet

224x224

32

arXiv

Inception v1

ImageNet

224x224

32

arXiv

Inception v2

ImageNet

224x224

32

arXiv

Inception v3

ImageNet

299x299

32

arXiv

Inception v4

ImageNet

299x299

32

arXiv

VGG 16

ImageNet

224x224

32

arXiv

VGG 19

ImageNet

224x224

32

arXiv

Detection architectures

Architecture

Backbone

Trained on

Input size

Batch size

Research paper

Yolo

v2

VOC 2007

416x416

64

arXiv

v3

ImageNet 2012

416x416

64

arXiv

Faster-RCNN

Resnet-50

COCO 2018

1024x1024*

1

arXiv

Resnet-101

COCO 2018

1024x1024*

1

arXiv

RFCN

Resnet-101

COCO 2018

1024x1024*

1

arXiv

SSD

Inception v2

COCO 2018

300x300

24

arXiv

MobileNet v1

COCO 2018

300x300

24

-

MobileNet v2

COCO 2018

300x300

24

-

SSDLite

MobileNet v2

COCO 2018

300x300

24

arXiv

* Faster-RCNN and RFCN don't require a fixed image input size. As such they can accomodate images from 600 to 1024 pixel.