Use Case and High-Level Description¶

This is a segmentation network to classify each pixel into four classes: BG, road, curb, mark.

Metric

Value

Image size

896x512

GFlops

4.770

MParams

0.184

Source framework

PyTorch*

Accuracy¶

The quality metrics calculated on 500 images from “Mighty AI” dataset that was converted for four class classification task are:

Label

IOU

ACC

mean

0.844

0.899

BG

0.986

0.994

0.954

0.974

curbs

0.727

0.825

marks

0.707

0.803

• IOU=TP/(TP+FN+FP)

• ACC=TP/GT

• TP - number of true positive pixels for given class

• FN - number of false negative pixels for given class

• FP - number of false positive pixels for given class

• GT - number of ground truth pixels for given class

Inputs¶

A blob with a BGR image and the shape 1, 3, 512, 896 in the format B, C, H, W, where:

• B – batch size

• C – number of channels

• H – image height

• W – image width

Outputs¶

The output is a blob with the shape 1, 4, 512, 896 in the format B, C, H, W. It can be treated as a four-channel feature map, where each channel is a probability of one of the classes: BG, road, curb, mark.

Demo usage¶

The model can be used in the following demos provided by the Open Model Zoo to show its capabilities: