This model is an instance segmentation network for 80 classes of objects. It is a Cascade mask R-CNN with ResNet101 backbone and deformable convolutions, FPN, RPN, detection and segmentation heads.
| Metric | Value |
|---|---|
| COCO val2017 box AP (max short side 800, max long side 1344) | 45.8% |
| COCO val2017 mask AP (max short side 800, max long side 1344) | 39.7% |
| COCO val2017 box AP (max height 800, max width 1344) | 43.55% |
| COCO val2017 mask AP (max height 800, max width 1344) | 38.14% |
| Max objects to detect | 100 |
| GFlops | 828.6324 |
| MParams | 101.236 |
| Source framework | PyTorch* |
Average Precision (AP) is defined and measured according to standard COCO evaluation procedure.
Image, name: image, shape: 1, 3, 800, 1344 in the format 1, C, H, W, where:
C- number of channelsH- image heightW- image width
The expected channel order is BGR
Model has outputs with dynamic shapes.
- Name:
labels, shape:-1- Contiguous integer class ID for every detected object. - 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]. - Name:
masks, shape:-1, 28, 28- Segmentation heatmaps for every output bounding box.
The OpenVINO Training Extensions provide a training pipeline, allowing to fine-tune the model on custom dataset.
The model can be used in the following demos provided by the Open Model Zoo to show its capabilities:
- Instance Segmentation Python* Demo
- Multi Camera Multi Target Python* Demo
- Whiteboard Inpainting Python* Demo
[*] Other names and brands may be claimed as the property of others.