Integrating Low-Level Visual Cues for Enhanced Unsupervised Semantic Segmentation
Authors: Yuhao Qing, Dan Zeng, Shaorong Xie, Kaer Huang, Yueying Wang
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on multiple datasets, including COCO-Stuff27, Cityscapes, Potsdam, and Ma STr1325, demonstrate that IL2Vseg achieves state-of-the-art results. Quantitative Evaluation. We report the results for the COCO-Stuff and Cityscapes datasets in Table 1 and Table 2. Ablation Study Low-Level Visual Cue. To further explore the proposed IL2Vseg, we conducted a series of ablation experiments. |
| Researcher Affiliation | Collaboration | 1School of Mechatronic Engineering and Automation, Shanghai University, Shanghai 200444, China 2School of Communication and Information Engineering, Shanghai University, Shanghai 200444, China 3Lenovo, Building 1, No.10 Xibeiwang East Road, Haidian District, Beijing, 100085, China |
| Pseudocode | No | The paper describes methods using equations and prose (e.g., "Our method involves: (1) Spatially Constrained Fuzzy Clustering, and (2) Colour Affinity for Feature Boosting."), but it does not contain a dedicated pseudocode block or algorithm section. |
| Open Source Code | No | The paper does not contain any explicit statement about releasing code, nor does it provide a link to a code repository. |
| Open Datasets | Yes | We use the COCOStuff(Caesar, Uijlings, and Ferrari 2018), Cityscapes(Cordts et al. 2016), Potsdam-3(Ji, Henriques, and Vedaldi 2019), and Mastr1325(Bovcon et al. 2019) datasets. |
| Dataset Splits | No | The paper mentions merging categories for COCOStuff and Cityscapes (resulting in 27 evaluation categories) and evaluating Potsdam-3 and Mastr1325 using three categories. It also states, "Consistent with existing methods, we use minibatch K-means based on cosine similarity for cluster segmentation." However, it does not explicitly specify training, validation, or test splits by percentage, sample count, or reference to predefined splits. |
| Hardware Specification | Yes | We conducted our experiments using the Py Torch 1.13 framework, running on an RTX 4090 GPU. |
| Software Dependencies | Yes | We conducted our experiments using the Py Torch 1.13 framework, running on an RTX 4090 GPU. |
| Experiment Setup | Yes | We used the Adam optimizer to optimize the nonlinear projections and local clustering centers, with learning rates set to 1 × 10−4 and 5 × 10−4, respectively... The weights β and γ are taken as 0.05 and 0.1 respectively. |