[Gen2] Deeplabv3 multiclass on DepthAI

This example shows how to run DeeplabV3+ on DepthAI in the Gen2 API system. The model has a MobilenetV2 backbone and is trained on Pascal VOC 2012 dataset. It contains the following classes:

• Person: person
• Animal: bird, cat, cow, dog, horse, sheep
• Vehicle: aeroplane, bicycle, boat, bus, car, motorbike, train
• Indoor: bottle, chair, dining table, potted plant, sofa, tv/monitor

You can find the tutorial for training the model and generation of .blob file here - DeepLabV3plus_MNV2.ipynb . You can create a copy of the Colab Notebook and try training the model on your own!

See the example in action detecting monitor and a person:

Pre-requisites

1. Purchase a DepthAI (or OAK) model (see shop.luxonis.com)
2. Install requirements

   python3 -m pip install -r requirements.txt
   

Usage

python3 main.py [options]

Options:

• -cam, --cam_input: Select camera input source for inference. Available options: left, right, rgb (default).
• -nn, --nn_model: Select model path for inference. Default: models/deeplab_v3_plus_mnv2_decoder_256_openvino_2021.4.blob