Skip to content

Latest commit

 

History

History
64 lines (44 loc) · 3.08 KB

File metadata and controls

64 lines (44 loc) · 3.08 KB

instance-segmentation-security-0002

Use Case and High-Level Description

This model is an instance segmentation network for 80 classes of objects. It is a Mask R-CNN with ResNet50 backbone, FPN, RPN, detection and segmentation heads.

Example

Specification

Metric Value
COCO val2017 box AP (max short side 768, max long side 1024) 40.8%
COCO val2017 mask AP (max short side 768, max long side 1024) 36.9%
COCO val2017 box AP (max height 768, max width 1024) 39.86%
COCO val2017 mask AP (max height 768, max width 1024) 36.44%
Max objects to detect 100
GFlops 423.0842
MParams 48.3732
Source framework PyTorch*

Average Precision (AP) is defined and measured according to standard COCO evaluation procedure.

Inputs

Image, name: image, shape: 1, 3, 768, 1024 in the format 1, C, H, W, where:

  • C - number of channels
  • H - image height
  • W - image width

The expected channel order is BGR.

Outputs

Model has outputs with dynamic shapes.

  1. Name: labels, shape: -1 - Contiguous integer class ID for every detected object.
  2. Name: boxes, shape: -1, 5 - Bounding boxes around every detected objects in (top_left_x, top_left_y, bottom_right_x, bottom_right_y) format and its confidence score in range [0, 1].
  3. Name: masks, shape: -1, 28, 28 - Segmentation heatmaps for every output bounding box.

Training Pipeline

The OpenVINO Training Extensions provide a training pipeline, allowing to fine-tune the model on custom dataset.

Demo usage

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

Legal Information

[*] Other names and brands may be claimed as the property of others.