Region Focus Network for Joint Optic Disc and Cup Segmentation
Authors: Ge Li, Changsheng Li, Chan Zeng, Peng Gao, Guotong Xie751-758
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results on the REFUGE-2018 challenge dataset and the Drishti-GS dataset show that the proposed method achieves the best performance, compared with competitive approaches reported in the literature and the official leaderboard. Furthermore, with much ablation study on the REFUGE1 dataset and the Drishti-GS (Sivaswamy et al. 2014) dataset, we evaluate the effectiveness of the proposed method, and the results demonstrate that our method achieves better performance, compared with the state-of-the-art. |
| Researcher Affiliation | Collaboration | 1Ping An Technology (Shenzhen) Co. Ltd., Shenzhen, China 2School of Computer Science and Technology, Beijing Institute of Technology |
| Pseudocode | No | The paper describes the network architecture and method in text and figures, but does not include any explicit pseudocode or algorithm blocks. |
| Open Source Code | No | Our code will be released soon. |
| Open Datasets | Yes | The first one is the REFUGE dataset, provided by the REFUGE-2018 challenge... 1https://refuge.grand-challenge.org/ The second one is the Drishti-GS dataset, provided by Medical Image Processing (MIP) group, IIIT Hyderabad. It contains 50 training images and 51 validation images. As the same in most works (Fu et al. 2018; Zhang et al. 2019), the dice coefficients of the optic cup (Dice OC) and disc (Dice OD), mean intersection-over-union (m Io U), as well as FPS (Frames Per Second), are employed as the evaluation metrics. ... Sivaswamy, J.; Krishnadas, S. R.; Joshi, G. D.; Jain, M.; and Tabish, A. U. S. 2014. Drishti-gs: Retinal image dataset for optic nerve head(onh) segmentation. 53 56. |
| Dataset Splits | Yes | The REFUGE-2018 challenge, including both normal and glaucomatous cases and ground-truth of segmentation from multiple human experts. The dataset consists of 1200 color fundus photographs which are split 1:1:1 into 3 subsets equally for training, offline validation and onsite test. ... The second one is the Drishti-GS dataset, provided by Medical Image Processing (MIP) group, IIIT Hyderabad. It contains 50 training images and 51 validation images. |
| Hardware Specification | Yes | In our experiments, we train the model on 1 NVIDIA Tesla P100 GPU for 100 epochs and employ stochastic gradient descent (SGD) for optimizing the deep model. |
| Software Dependencies | No | The paper mentions using FPN with Mobile Net-v1 as the backbone but does not specify software dependencies like programming language versions or library versions (e.g., Python version, TensorFlow/PyTorch version). |
| Experiment Setup | Yes | Images are resized and padded with zeros to get a square image such that their scale is 512 pixels. Each mini-batch has 4 images per GPU and each image has N sampled Ro Is, with a ratio of 1:3 of positives to negatives. N is set to 200. In our experiments, we train the model on 1 NVIDIA Tesla P100 GPU for 100 epochs and employ stochastic gradient descent (SGD) for optimizing the deep model. We use a gradually decreasing learning rate starting from 0.01 and a momentum of 0.9. We employ the piecewise constant learning rate policy where the learning rate is multiplied by 0.1 every 30 epoch. |