A Continual Learning Framework for Uncertainty-Aware Interactive Image Segmentation
Authors: Ervine Zheng, Qi Yu, Rui Li, Pengcheng Shi, Anne Haake6030-6038
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments are performed on both medical and natural image datasets to illustrate the proposed framework s effectiveness on basic segmentation performance, forward knowledge transfer, and backward knowledge transfer. |
| Researcher Affiliation | Academia | Ervine Zheng, Qi Yu *, Rui Li, Pengcheng Shi, Anne Haake Rochester Institute of Technology {mxz5733, qi.yu, rxlics, spcast, arhics}@rit.edu |
| Pseudocode | No | The paper does not contain any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The source code and the appendix are available at https://github.com/ritmininglab/CLIS |
| Open Datasets | Yes | The first dataset is Cityscapes (Cordts et al. 2016) containing street scenes from 50 different cities, with pixel-level annotations of 5000 frames and 8 major semantic categories. The second dataset is Ma STr1325 (Bovcon et al. 2019) for marine semantic segmentation and obstacle detection. The third dataset is from the ISIC challenge for skin lesion analysis towards melanoma detection (Codella et al. 2018) with dermoscopic lesion images. |
| Dataset Splits | No | The paper defines a training-testing protocol for continual learning tasks (basic performance, forward knowledge transfer, and backward knowledge transfer) rather than a fixed train/validation/test split for a single dataset in a conventional manner. |
| Hardware Specification | No | The paper does not specify any hardware details such as GPU models, CPU types, or memory used for the experiments. |
| Software Dependencies | No | The paper mentions using 'Adam optimizer' and 'Slic algorithm' but does not provide specific version numbers for any software libraries or dependencies. |
| Experiment Setup | Yes | The hyper-parameters are set to α0 = 1, β0 = 1, µ0 = 0, σ0 = 1, τ = 2. We use the Slic algorithm (Achanta et al. 2012) to group visually-similar pixels into superpixels and reduce the computational cost. The Adam optimizer is used for gradientbased model updates once user annotations are collected. |