Use Case and High-Level Description¶

This model presents a person attributes classification algorithm analysis scenario. It produces probability of person attributions existing on the sample and a position of two point on sample, which can be used for color prob (like, color picker in graphical editors)

Specification¶

Metric

Value

Pedestrian pose

Standing person

Occlusion coverage

<20%

Min object width

80 pixels

Supported attributes

is_male, has_bag, has_backpack, has hat, has longsleeves, has longpants, has longhair, has coat_jacket

GFlops

0.174

MParams

0.735

Source framework

PyTorch*

Accuracy¶

Attribute

F1

is_male

0.91

has_bag

0.66

has_backpack

0.77

has_hat

0.64

has_longsleeves

0.21

has_longpants

0.83

has_longhair

0.83

has_coat_jacket

NA

Inputs¶

Image, name: input, shape: 1, 3, 160, 80 in the format 1, C, H, W, where:

• C - number of channels

• H - image height

• W - image width

The expected color order is BGR.

Outputs¶

1. The net outputs a blob named 453 with shape: 1, 8, 1, 1 across eight attributes: [is_male, has_bag, has_backpack, has_hat, has_longsleeves, has_longpants, has_longhair, has_coat_jacket]. Value > 0.5 means that an attribute is present.

2. The net outputs a blob named 456 with shape: 1, 2, 1, 1. It is location of point with top color.

3. The net outputs a blob named 459 with shape: 1, 2, 1, 1. It is location of point with bottom color.

Demo usage¶

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