Quality Evaluation of Image Dataset based on Label File
8th International Conference on Dependable Systems and Their Applications, DSA 2021
; : 639-646, 2021.
Article
in English
| Scopus | ID: covidwho-1672601
ABSTRACT
The quality of the dataset affects the accuracy of the artificial intelligence model, but it is a lot of work to manually detect errors related to the quality evaluation of the dataset, and it may not be possible to perform quality evaluation through simple viewing. Therefore, we propose an image dataset quality measurement model, including nine evaluation metrics, and analyze the evaluation metrics from three aspects definition, calculation formula and description. Based on the label file, the quality of the dataset file and the content of the dataset is evaluated, and the evaluation standard is given to judge whether the quality of the dataset is qualified. The measurement model and evaluation criteria proposed in this article were verified against the Cifar-10 dataset and the COVID-CT dataset, and the problems of label accuracy and label category imbalance were found, which proved the effectiveness of the method in this paper. © 2021 IEEE.
Full text:
Available
Collection:
Databases of international organizations
Database:
Scopus
Type of study:
Experimental Studies
Language:
English
Journal:
8th International Conference on Dependable Systems and Their Applications, DSA 2021
Year:
2021
Document Type:
Article
Similar
MEDLINE
...
LILACS
LIS