SVasP: Self-Versatility Adversarial Style Perturbation for Cross-Domain Few-Shot Learning
Authors: Wenqian Li, Pengfei Fang, Hui Xue
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on multiple benchmark datasets demonstrate that our method significantly outperforms existing state-of-the-art methods. Experiments Datasets Following the BSCD-FSL benchmark proposed in BSCD-FSL (Guo et al. 2020) and the mini-CUB benchmark proposed in FWT (Tseng et al. 2020), we use mini Image Net (Ravi and Larochelle 2017) with 64 classes as the source domain. The target domains include eight datasets: Chest X (Wang et al. 2017), ISIC (Tschandl, Rosendahl, and Kittler 2018), Euro SAT (Helber et al. 2019), Crop Disease (Mohanty, Hughes, and Salath e 2016), CUB (Wah et al. 2011), Cars (Krause et al. 2013), Places (Zhou et al. 2017), and Plantae (Van Horn et al. 2018). |
| Researcher Affiliation | Academia | 1School of Computer Science and Engineering, Southeast University, Nanjing 210096, China 2Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, China EMAIL |
| Pseudocode | Yes | More construction details and the complete adversarial style generation pseudo-code can be found in Appendix A. |
| Open Source Code | Yes | Code https://github.com/liwenqianSEU/SVasP |
| Open Datasets | Yes | Following the BSCD-FSL benchmark proposed in BSCD-FSL (Guo et al. 2020) and the mini-CUB benchmark proposed in FWT (Tseng et al. 2020), we use mini Image Net (Ravi and Larochelle 2017) with 64 classes as the source domain. The target domains include eight datasets: Chest X (Wang et al. 2017), ISIC (Tschandl, Rosendahl, and Kittler 2018), Euro SAT (Helber et al. 2019), Crop Disease (Mohanty, Hughes, and Salath e 2016), CUB (Wah et al. 2011), Cars (Krause et al. 2013), Places (Zhou et al. 2017), and Plantae (Van Horn et al. 2018). |
| Dataset Splits | Yes | Specifically, to simulate the N-way K-shot problem, N classes are selected and K samples per class are chosen to form the support set S = {xs i, ys i }ns i=1, where ns = NK. And the same N classes with another M images are used to construct the query set Q = {xq i }nq i=1, where nq = NM. Therefore, an episode T = (S, Q) is constituded, comprising of a support set S and a query set Q, and |T | = N(K + M). ... Each class contains 5 support samples and 15 query samples. |
| Hardware Specification | Yes | All the experiments are conducted on a single NVIDIA GeForce RTX 3090. |
| Software Dependencies | No | The paper mentions using Res Net-10 and Vi T-small as backbones, GNN and Proto Net as classifiers, and Adam/SGD as optimizers, but does not provide specific version numbers for any software libraries or frameworks. |
| Experiment Setup | Yes | The optimizer is Adam with a learning rate of 0.001. ... The network is meta-trained for 200 epochs with 120 episodes per epoch. ... The optimizer is SGD with a learning rate of 5e-5 and 0.001 for E and fre, respectively. ... Hyperparameters are set as follows: ξ = 0.1, k = 2, λ = 0.2 and choose κ1, κ2 from [0.008, 0.08, 0.8]. The probability to perform style change is set to 0.2. |