Supported NNs

Supervisely suppports most of the state of the art models for common computer vision tasks:

• Interactive segmentation

• Semantic segmentation

• Instance segmentation

• Object Detection

• Classification

• Text detection

• OCR

Training, Inference, Pre-trained weights : off the shelf

All neural networks architectures (listed below) support both training and inference inside the Supervisely Platform. Also we provide pretrained weights for each architecture that can be used directly for inference or for transfer learning to speed up the training process on your custom data.

Source codes and customization

We aslo provide source codes for all neural networks so developers can customize them. Read more in the SDK chapter.

Interactive segmentation

Architecture	Pretrained on	Framework	Paper
Smart Tool	Custom data	PyTorch	Medium
	Custom data + Supervisely Person		Medium1, Medium2

Semantic segmentation

Architecture	Pretrained on	Framework	Paper
Unet V2 (based on VGG)	ImageNet	PyTorch	arxiv
PSPNet	Cityscapes	Tensorflow	arxiv
	ADE20K
ICNet	Cityscapes	Tensorflow	arxiv
	ADE20K
DeepLab V3 +	Cityscapes	Tensorflow	arxiv
	ADE20K

Instance segmentation

Architecture	Backbone	Pretrained on	Framework	Paper
Mask-RCNN	Inception V2	COCO	Tensorflow	arxiv
	ResNet-50 (atrous)	COCO
	ResNet-101 (atrous)	COCO

Object detection

Architecture	Backbone	Pretrained on	Framework	Paper
YOLO V3		COCO	DarkNet (C++)	arxiv
SSD	Inception V2	COCO	Tensorflow	arxiv
	MobileNet V1	COCO
	MobileNet V2	COCO
	MobileNet V2 (lite)	COCO
Faster-RCNN	ResNet-50	COCO	Tensorflow	arxiv
	ResNet-101	COCO
	Inception V2	COCO
	Inception ResNet V2 (atrous)	COCO
	NasNet	COCO
	NasNet (lowproposals)	COCO
RFCN	ResNet-101	COCO	Tensorflow	arxiv

Classification

Architecture	Backbone	Pretrained on	Framework	Paper
AlexNet		ImageNet	PyTorch	arxiv
VGG	*-11	ImageNet	PyTorch	arxiv
	*-13	ImageNet
	*-16	ImageNet
	*-19	ImageNet
	*-11 with batchnorm	ImageNet
	*-13 with batchnorm	ImageNet
	*-16 with batchnorm	ImageNet
	*-19 with batchnorm	ImageNet
ResNet	*-18	ImageNet	PyTorch	arxiv
	*-34	ImageNet
	*-50	ImageNet
	*-101	ImageNet
	*-152	ImageNet
SqueezeNet	*-1.0	ImageNet	PyTorch	arxiv
	*-1.1	ImageNet
Densenet	*-121	ImageNet	PyTorch	arxiv
	*-169	ImageNet
	*-201	ImageNet
	*-161	ImageNet
Inception V3		ImageNet	PyTorch	arxiv

Text detection

Architecture	Pretrained on	Framework	Paper
EAST	ICDAR 2015	Tensorflow	arxiv
CTPN	ICDAR 2015	Tensorflow	arxiv
PixelLink	ICDAR 2015	Tensorflow	arxiv

OCR

Architecture	Pretrained on	Framework	Paper
CNN-LSTM-CTC	ICDAR 2015	Tensorflow	arxiv