Enhancing Generalized Few-Shot Semantic Segmentation via Effective Knowledge Transfer
Authors: Xinyue Chen, Miaojing Shi, Zijian Zhou, Lianghua He, Sophia Tsoka
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on PASCAL-5i and COCO-20i demonstrate that our approach significantly enhances the state of the art in the GFSS setting. Table 1 and 2 present a comparison of several GFSS methods with our method on the PASCAL-5i and COCO-20i datasets. We conduct extensive experiments on PASCAL-5i and COCO-20i, which demonstrate that our method effectively improves previous GFSS approaches. |
| Researcher Affiliation | Academia | 1Department of Informatics, King s College London, UK 2 College of Electronic and Information Engineering, Tongji University, China EMAIL, EMAIL |
| Pseudocode | No | The paper describes the methods in detailed paragraphs and uses figures (e.g., Figure 2 for method overview) to illustrate the architecture and process. However, it does not include any explicit pseudocode blocks or algorithms. |
| Open Source Code | Yes | Code https://github.com/xinyue1chen/GFSS-EKT |
| Open Datasets | Yes | We evaluate our model on two public datasets, PASCAL-5i (Shaban et al. 2017) and COCO-20i (Nguyen and Todorovic 2019). |
| Dataset Splits | Yes | Object categories in each dataset are evenly split into four folds. Following (Tian et al. 2022), we adopt a cross-validation manner to train the model on three folds while testing on one fold. This procedure is repeated four times, and we report the average result. We perform K {1, 5} shot semantic segmentation. |
| Hardware Specification | Yes | Our method is implemented in Py Torch with NVIDIA A100. |
| Software Dependencies | No | Our method is implemented in Py Torch with NVIDIA A100. We utilize PSPNet (Zhao et al. 2017) with Res Net50 (He et al. 2016) as the feature extractor. The paper mentions Py Torch but does not specify its version, nor does it list versions for other software dependencies or libraries. |
| Experiment Setup | Yes | In the pre-training phase, the model is optimized using stochastic gradient descent (SGD) with an initial learning rate of 0.01, a momentum of 0.9, and a weight decay of 0.0001. The model is trained for 50 epochs on both datasets, and the batch size is set as 8 and 16 for PASCAL-5i and COCO-20i, respectively. In the fine-tuning phase, we update the model using SGD with a learning rate of 0.01, training for 500 epochs on both datasets. |