Self-Disentanglement and Re-Composition for Cross-Domain Few-Shot Segmentation
Authors: Jintao Tong, Yixiong Zou, Guangyao Chen, Yuhua Li, Ruixuan Li
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments show that our model outperforms the state-of-the-art CD-FSS method by 1.92% and 1.88% in average accuracy under 1-shot and 5-shot settings, respectively. Table 2: Mean-IoU of 1-shot and 5-shot results on the CD-FSS benchmark. The best and second-best results are highlighted in bold and underlined, respectively. Table 3: Detailed ablation study results of our various designs on four target datasets under 1-shot setting and 5-shot setting. |
| Researcher Affiliation | Academia | 1School of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan, China 2School of Computer Science, Peking University, Beijing, China. |
| Pseudocode | No | The paper describes the methodology using mathematical equations and textual descriptions in sections 2 and 3, but does not include a clearly labeled pseudocode or algorithm block. |
| Open Source Code | No | The paper does not contain an explicit statement about releasing code or a link to a code repository. |
| Open Datasets | Yes | PASCAL VOC 2012 (Everingham et al., 2010) with SBD (Hariharan et al., 2011) augmentation serves as the training dataset. FSS-1000 (Li et al., 2020), Deep Globe (Demir et al., 2018), ISIC2018 (Codella et al., 2019; Tschandl et al., 2018), and Chest X-ray (Candemir et al., 2013; Jaeger et al., 2013) are considered as target domains for evaluation. |
| Dataset Splits | No | The paper states: 'We adopt the meta-learning episodic manner following (Lei et al., 2022) to train and test our model. Specifically, both the training set from Ds and the testing set from Dt consist of several episodes.' It also mentions 'official dataset split' for FSS-1000 and 'standards established by PATNet' for ISIC2018, but does not provide explicit train/validation/test percentages or sample counts for all datasets used in its experiments beyond the episodic description. |
| Hardware Specification | No | The paper states: 'The computation is completed in the HPC Platform of Huazhong University of Science and Technology.' This is a general statement and does not provide specific hardware details such as GPU/CPU models or memory. |
| Software Dependencies | No | The paper does not provide specific software dependency details such as library names with version numbers (e.g., Python 3.8, PyTorch 1.9, CUDA 11.1). |
| Experiment Setup | Yes | The hyperparameter λ, which adjusts the weight of the orthogonal loss, is set to 0.1, and the rank r of the OSD module is set to 8. During both source-domain training and target-domain fine-tuning, we employ the standard Binary Cross-Entropy (BCE) loss LBCE, with the orthogonal loss Lorth from OSD added as a regularization term to promote semantic decoupling. We built a lightweight encoder-only baseline that employs ViT-B pre-trained on ImageNet as the backbone network. |