SFUOD: Source-Free Unknown Object Detection (ICCV 2025)

Keon-Hee Park, Seun-An Choe, Gyeong-Moon Park

Official Pytorch implementation of SFUOD: Source-Free Unknown Object Detection, ICCV 2025.

1. Installation

1.1 Requirements

• Linux, CUDA >= 11.1, GCC >= 8.4

• Python >= 3.8

• torch >= 1.10.1, torchvision >= 0.11.2

• Other requirements

  pip install -r requirements.txt

1.2 Compiling Deformable DETR CUDA operators

cd ./models/ops
sh ./make.sh
# unit test (should see all checking is True)
python test.py

2. Dataset Preparation

Weather Adaptation:

Cityscapes (source domain) → FoggyCityscapes with foggy level 0.02 (target domain).

You can download the raw data from the official websites: Cityscapes, FoggyCityscapes. The annotations, converted into COCO format, can download from here. The datasets and annotations are organized as:

[data_root]
└─ cityscapes
	└─ annotations
		└─ cityscapes_train_cocostyle.json
		└─ cityscapes_val_cocostyle.json
	└─ leftImg8bit
		└─ train
		└─ val
└─ foggy_cityscapes
	└─ annotations
		└─ foggy_cityscapes_train_cocostyle.json
		└─ foggy_cityscapes_val_cocostyle.json
	└─ leftImg8bit_foggy
		└─ train
		└─ val

3. Training

first edit the files in configs/def-detr-base/city2foggy/ to specify your own DATA_ROOT and OUTPUT_DIR, then run:

3.1 Source Training

sh configs/def-detr-base/city2foggy/source_only_sfuod.sh

3.2 Target Training

Mean-Teacher framework

sh configs/def-detr-base/city2foggy/teaching_standard_sfuod.sh

DRU (ECCV 2024)

sh configs/def-detr-base/city2foggy/teaching_mask_sfuod.sh

Target Training (Oracle)

sh configs/def-detr-base/city2foggy/target_only_sfuod.sh

4. Acknowledgement

This implementation is built upon DRU, Deformable DETR, and MIC. We sincerely appreciate their contributions.