Xuanlong/FT-FSOD-CD-FSOD

FT-FSOD / CD-FSOD Model Checkpoints

This repository provides the Swin-B checkpoints used to reproduce the results in A Closer Look at Cross-Domain Few-Shot Object Detection: Fine-Tuning Matters and Parallel Decoder Helps. Checkpoints are available for six target datasets under the 1-shot, 5-shot, and 10-shot settings.

Model Details

• Task: Cross-domain few-shot object detection
• Backbone: Swin-B
• Framework: MMDetection-compatible checkpoint format
• Target datasets: ArTaxOr, Clipart1k, DIOR, DeepFish, NEU-DET, and UODD
• Few-shot settings: 1, 5, and 10 shots
• Purpose: Reproducing the results reported in the paper

The checkpoints do not contain optimizer state or training-state metadata. They require the matching model definition and experiment configuration to construct the detector before loading the weights.

Results and Checkpoints

The table reports mAP on the CD-FSOD benchmark. Each score links to the checkpoint used to produce it.

Shots	ArTaxOr	Clipart1k	DIOR	DeepFish	NEU-DET	UODD	Avg.
1	49.1	55.6	24.6	42.7	15.5	22.1	34.9
5	76.8	59.4	35.3	45.5	25.2	27.5	45.0
10	79.2	59.6	41.5	46.3	28.4	32.3	47.9

We also encourage training the models on your own machine. Because each target domain contains very few training examples, results can vary slightly between runs. Some datasets may perform better than the reported results, while others may perform slightly worse. With the same data splits and configuration, the overall variation is expected to remain relatively small.

Download

Download the complete repository with the Hugging Face CLI:

hf download Xuanlong/FT-FSOD-CD-FSOD --local-dir FT-FSOD-CD-FSOD

Download one checkpoint by specifying its repository path:

hf download Xuanlong/FT-FSOD-CD-FSOD \
  swinB_all_DIOR_1shot/best_coco_bbox_mAP_iter_170.pth \
  --local-dir FT-FSOD-CD-FSOD

License

License information will be added before the final release. Until then, no license is granted for use, modification, or redistribution beyond rights provided by applicable law.

Citation

@inproceedings{yu2026closer,
  title={A closer look at cross-domain few-shot object detection: Fine-tuning matters and parallel decoder helps},
  author={Yu, Xuanlong and Sha, Youyang and Liu, Longfei and Shen, Xi and Yang, Di},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={26593--26603},
  year={2026}
}