📝 Small Multi-Object Tracking for Spotting Birds (SMOT4SB) Challenge 2025

---

🎛 Dependencies

---

• Install conda environment for both training and inference

conda create -n smot4sb python=3.11
conda activate smot4sb
python3 -m pip install -r requirements.txt
python3 -m pip install -r requirements_mmdet.txt

Download data

---

• [FBD dataset](FBD-SV-2024: Flying Bird Object Detection Dataset in Surveillance Video): This external dataset is used together with S0D4SB to train bird detectors.

# figshare link
https://figshare.com/s/1ca0193680f894a65371?file=50113731

# Repos:
https://github.com/Ziwei89/FBD-SV-2024_github.git
https://github.com/Ziwei89/FBOD

# Command
wget https://figshare.com/ndownloader/files/50113731?private_link=1ca0193680f894a65371 -O ./data/FBD-SV-2024.zip

• [SOD4SB](MVA2023 Small Object Detection Challenge for Spotting Birds: Dataset, Methods, and Results)

https://drive.google.com/drive/u/1/folders/1WnvpWi8C7GHu_OtXu8DHhDhrk3fCUliI

gdown https://drive.google.com/drive/u/1/folders/1cPscYgFrBiuYqmN1U2QXyh4ewWiwxUXw -O  ./data/sod4sb --folder --continue --remaining-ok

How to Run Video tracking

• Download trained weights:
  • Download model weights from the following Google Drive Url
  • Unzip the downloaded zip files into the following structure:

checkpoints/model_weights/
├── bird_classification_mambaout.pth
├── centernet_efficientnet_merged_dataset_best_coco_bbox_mAP_epoch_31.pth
├── centernet_efficientnet_smot4sb_best_coco_bbox_mAP_epoch_17.pth
└── centernet_rexnet_150_merged_dataset_best_coco_bbox_mAP_epoch_36.pth

• Run the tracking on each video with the following script:

python3 scripts/track_video.py --video_dir /path/to/video/video_name

• The output tracking result for each frame will be saved into predictions/pub_test/{video_name}

💎 References

---

[1] MVA2023 - Small Object Detection Challenge for Spotting Birds

[2] FBD-SV-2024: Flying Bird Object Detection Dataset in Surveillance Video