Deep Neural Networks Learn Meta-Structures from Noisy Labels in Semantic Segmentation

Authors: Yaoru Luo, Guole Liu, Yuanhao Guo, Ge Yang1908-1916

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To quantitatively analyze the segmentation performance of DNNs trained by these labels, we experiment on two representative segmentation models, U-Net (Ronneberger, Fischer, and Brox 2015) and Deep Labv3+ (Chen et al. 2018), with the same loss function (binary cross-entropy loss) and optimizer (stochastic gradient descent, SGD).
Researcher Affiliation Academia 1National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences 2School of Artificial Intelligence, University of Chinese Academy of Sciences
Pseudocode Yes Algorithm 1: Unsupervised Iteration Strategy
Open Source Code No The paper does not provide any explicit statement or link indicating that the source code for the methodology described is publicly available.
Open Datasets Yes For binary-class segmentation, we select fluorescence microscopy images of ER, MITO datasets (Luo, Guo, and Yang 2020) and the NUC dataset (Caicedo et al. 2019). For multi-class segmentation, we select natural images of Cityscapes dataset (Cordts et al. 2016).
Dataset Splits No The paper mentions 'testing dice scores during training' but does not explicitly provide specific training/validation/test dataset splits (e.g., percentages, counts, or detailed methodology) in the main text. It refers to Appendix C for 'Detailed information on the datasets and experimental configurations', but this information is not directly in the main body.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions the use of U-Net and Deep Labv3+ models, binary cross-entropy loss, and stochastic gradient descent (SGD) optimizer, but does not provide specific version numbers for any software dependencies or libraries.
Experiment Setup No The paper mentions using U-Net and Deep Labv3+ models, binary cross-entropy loss, and SGD optimizer. It states that 'Detailed information on the datasets and experimental configurations are provided in Appendix C.', implying hyperparameters are not detailed in the main text.