[paper]

To access the main codebase, please switch to the tinypedseg_code branch.

To run the code in this repository, you need to create a Conda environment with the following dependencies:

conda create -n tos_env python=3.9 -y
conda activate tos_env

pip install torch==2.0.0 torchvision==0.15.1 mmsegmentation==0.27.0 mmengine==0.10.1
pip install mmcv-full==1.6.2 -f https://download.openmmlab.com/mmcv/dist/cu117/torch2.0.0/index.html

Ensure you have CUDA 11.7 installed and configured correctly. The required CUDA dependencies should be included with the above packages, but verify with:

nvcc --version

To train a model, use the following command structure:

python tools/train.py --config <path_to_config> --work-dir <path_to_output_directory>

For training with the deeplabv3plus_resnet_tos configuration:

python tools/train.py --config ./configs/tos/deeplabv3plus_resnet_tos.py --work-dir ./work_dirs/deeplabv3plus_resnet

This repository includes multiple model configurations:

• configs/tos/deeplabv3_unet_tos.py
• configs/tos/deeplabv3plus_resnet_tos.py
• configs/tos/swin_l_tos.py
• configs/tos/unet_tos.py

The user should select one of the available configuration files and specify the working directory accordingly.

• Ensure that you have a compatible GPU with CUDA 11.7 support.
• If additional dependencies are required, install them using pip install <package_name>.
• For further details, refer to the official documentation of the libraries used:
  • PyTorch
  • Torchvision
  • MMSegmentation
  • MMCV
  • MMEngine

If you find our project helpful, please consider citing our work.

@inproceedings{sahin2023tinypedseg,
  title={TinyPedSeg: A Tiny Pedestrian Segmentation Benchmark for Top-Down Drone Images},
  author={Sahin, Yusuf H and Abdinli, Elvin and Ayd{\i}n, M Arda and Unal, Gozde},
  booktitle={2023 18th International Conference on Machine Vision and Applications (MVA)},
  pages={1--5},
  year={2023},
  organization={IEEE}
}

This implementation builds upon MMSegmentation. We gratefully acknowledge their valuable contributions.