Mask R-CNN on DepthAI

This example shows how to run a Mask R-CNN model on DepthAI. Please note that since it is a heavier model so the expected speed is ~ 2-4 FPS

Export

You can export your Mask R-CNN directly to OpenVINO IR and then to blob. For the first conversion you can use this repository, where you would select torchvision.models.detection.mask_rcnn.maskrcnn_resnet50_fpn model and load your weights. More instruction can be found in the repository. After obtaining XML and BIN you can use blobconverter to convert them to blob and use them on OAK.

Pre-requisites

Install requirements:

python3 -m pip install -r requirements.txt

Usage

python3 main.py [options]

Options:

• -t or --threshold: Threshold for filtering out detections with lower probability. Default 0.5.
• -rt or --region_threshold: Threshold for filtering out mask points with low probability. Default: 0.5.