We participated in the 2023 IEEE Data Fusion Contest as a team known as "RoofDetective," an annual event. This year's competition featured two distinct tracks. Our team took part in Track 1, which focused on building detection and roof type classification. For more information on our approach, you can access our detailed report through this link. We ranked 21st among approximately 200 teams across the world.
- Assoc. Prof. Dr. Erchan Aptoula (Sabanci University)
- Efkan Durakli (Gebze Technical University)
- Deren Ege Turan (Sabanci University)
- Ekin Beyazit (Sabanci University)
- Emirhan Böge (Sabanci University)
Step 1. clone repository and update submodules
git clone https://github.com/efkandurakli/RoofDetective-DFC-2023.git
cd RoofDetective-DFC-2023
git submodule update --init --remote --recursive
Step 2. create conda virtual environment and activate it
conda create --name roof-detective-dfc2023 python=3.8
conda activate roof-detective-dfc2023
Step 3. install Pytorch following official instructions
Step 4. Install MMCV using MIM
pip install -U openmim
mim install mmcv-full
Step 5. Install pycocotools, mmengine and MMDetection
conda install -c conda-forge pycocotools
pip install mmengine
cd mmdetection
pip install -v -e .
Step 6. Install other required packages
pip install future tensorboard
conda install -c conda-forge tqdm
Step 7. Download config and checkpoint files and verify your installation
mim download mmdet --config yolov3_mobilenetv2_320_300e_coco --dest .
python demo/image_demo.py demo/demo.jpg yolov3_mobilenetv2_320_300e_coco.py yolov3_mobilenetv2_320_300e_coco_20210719_215349-d18dff72.pth --device cpu --out-file result.jpg
You will see a new image result.jpg on your current folder, where bounding boxes are plotted on cars, benches, etc.
python tools/train.py $CONFIG --work-dir $CHECKPOINT_DIR
python tools/test.py $CONFIG $checkpoint --format-only --eval-options "jsonfile_prefix=$SAVE_PATH"