Real-time Drone Instance Segmentation in Videos Using Optimized Mask Regional-Convolutional Neural Network

Installation

Required Libraries (Version)

detectron2 == 0.6
torch 2.0.1+cu118
NVCC 11.5

Install detectron2 (Please use WSL)

python -m pip install 'git+https://github.com/facebookresearch/detectron2.git'

Install torch

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

Anti-UGV-IS Dataset

Google drive

Training

• Download the annotated dataset and create split using "04_split_dataset.py".
• To run hyperparameter optimization run "05_hyperparameter_optimization.py". Note that some backbone may took 11GB+ vram. So its better to use google colab (Tesla T4).
• After that run "06_single_train.py". The model will be saved in the "Output" folder. That can be use for inferencing.

Inferencing