Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
Deconfound Semantic Shift and Incompleteness in Incremental Few-shot Semantic Segmentation
Authors: Yirui Wu, Yuhang Xia, Hao Li, Lixin Yuan, Junyang Chen, Jun Liu, Tong Lu, Shaohua Wan
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on both PASCAL-VOC 2012 and ADE20k benchmarks demonstrate the outstanding performance of our method. |
| Researcher Affiliation | Academia | 1Key Laboratory of Water Big Data Technology of Ministry of Water Resources, Hohai University, Nanjing, China 2College of Computer Science and Software Engineering, Shenzhen University, Shenzhen, China 3School of Computing and Communication, Lancaster University, Lancaster, UK 4National Key Lab for Novel Software Technology, Nanjing University, Nanjing, China 5Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China, Shenzhen, China |
| Pseudocode | No | The paper describes methods (CIM and PRM) using text and flow diagrams (Figures 3, 4, 5) but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any statement or link indicating that the source code for the methodology is openly available. |
| Open Datasets | Yes | Datasets. We conduct experiments on two widely used ISS datasets, i.e., Pascal-VOC 2012 (Everingham et al. 2015) and ADE20k (Zhou et al. 2017). |
| Dataset Splits | Yes | IFSS settings. 1) Tasks. For Pascal-VOC 2012, we consider three tasks, including 19-1 (T=2), 15-5 (T=2), and 15-1 (T=6). For ADE20k, we similarly consider three tasks, including 100-50 (T=2), 50-50 (T=3), and 100-10 (T=6). 2) Disjoint or overlapped. As described in (Cermelli et al. 2020a), images only contain pixels of the previous or current classes in the disjoint settings. In the overlapped settings, images may contain pixels of future classes, which presents a more realistic and challenging scenario, leading us to conduct experiments under such setting. |
| Hardware Specification | No | The paper mentions "High Performance Computing Platform, Hohai University" in the acknowledgments, but does not provide specific hardware models (e.g., GPU/CPU models, memory details). |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers. |
| Experiment Setup | Yes | Table 5 shows the ablation results with different λ and Lr values, where λ is the weight of the distillation loss, Lr is the learning rate for Step 0, and the learning rate for further steps is 0.1*Lr. We use λ = 10 for all experiments for a fair comparison. |