MiniSeg: An Extremely Minimum Network for Efficient COVID-19 Segmentation
Authors: Yu Qiu, Yun Liu, Shijie Li, Jing Xu4846-4854
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To address the above problems, we propose Mini Seg, a lightweight deep learning model for efficient COVID-19 segmentation. Compared with traditional segmentation methods, Mini Seg has several significant strengths: i) it only has 83K parameters and is thus not easy to overfit; ii) it has high computational efficiency and is thus convenient for practical deployment; iii) it can be fast retrained by other users using their private COVID-19 data for further improving performance. In addition, we build a comprehensive COVID-19 segmentation benchmark for comparing Mini Seg to traditional methods. Experiments demonstrate that Mini Seg performs favorably against previous state-of-the-art segmentation methods with high efficiency, trained with limited COVID-19 data. |
| Researcher Affiliation | Academia | Yu Qiu1, Yun Liu2*, Shijie Li3, Jing Xu1* 1 College of Artificial Intelligence, Nankai University 2 College of Computer Science, Nankai University 3 Department of Information Systems and Artificial Intelligence, University of Bonn yqiu@mail.nankai.edu.cn, {vagrantlyun, jason.li.nku}@gmail.com, xujing@nankai.edu.cn |
| Pseudocode | No | The paper describes the proposed methods using mathematical formulas and descriptive text, but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper states: "The code of these methods is provided online by the authors." referring to other state-of-the-art methods it compares against, but does not provide an explicit link or statement about the open-source availability of Mini Seg's own code. |
| Open Datasets | Yes | We utilize four open-access COVID-19 CT segmentation datasets, i.e., two sub-datasets from COVID19 CT Segmentation Dataset (Jenssen 2020), COVID-19 CT Lung and Infection Segmentation Dataset (Jun et al. 2020), and Mos Med Data (Morozov et al. 2020), to evaluate Mini Seg. According to the number of CT slices or the number of COVID-19 patients, we rename these datasets as COVID-19-CT100, COVID-19-P9, COVID-19-P20, and COVID-19-P1110 for convenience, respectively. The information of these datasets is summarized in Tab. 1. |
| Dataset Splits | Yes | Moreover, we perform 5-fold cross-validation to avoid statistically significant differences in performance evaluation. |
| Hardware Specification | Yes | Moreover, we also report the number of parameters, the number of FLOPs, and speed, tested using a 512 512 input image and a TITAN RTX GPU. |
| Software Dependencies | No | We implement the proposed Mini Seg network using the well-known Py Torch framework (Paszke et al. 2017). Adam optimization (Kingma and Ba 2015) is used for training with the weight decay of 1e-4. The paper mentions Py Torch and Adam but does not specify their version numbers. |
| Experiment Setup | Yes | We implement the proposed Mini Seg network using the well-known Py Torch framework (Paszke et al. 2017). Adam optimization (Kingma and Ba 2015) is used for training with the weight decay of 1e-4. We adopt the learning rate policy of poly, where the initial learning rate is 1e-3. We train 80 epochs on the training set with a batch size of 5. |