GSoc/2022/StatusReports/PhuocKhanhLe: Difference between revisions
| (3 intermediate revisions by the same user not shown) | |||
| Line 2: | Line 2: | ||
The main idea of IQS in digiKam is to determine the quality of an image and convert it into a score. This score is based on four factors sabotaging image: blur, noise, exposure, and compression. The current approach helps determine whether images are distorted for one of these reasons. However, the current algorithm also presents some drawbacks: It demands lots of fine-tuning from the user’s side and cannot work on the aesthetic image. So, I propose the solution of the deep learning algorithm. While the dataset and the paper for aesthetic image quality assessment are free to use, we are capable of constructing a mathematical model that can learn the pattern of a dataset, hence, predicting the score of quality. As deep learning is an end-to-end solution, it doesn’t require the setting for the hyperparameter. Therefore, we can reduce most of the fine-tuning parts to make this feature easier to use | The main idea of IQS in digiKam is to determine the quality of an image and convert it into a score. This score is based on four factors sabotaging image: blur, noise, exposure, and compression. The current approach helps determine whether images are distorted for one of these reasons. However, the current algorithm also presents some drawbacks: It demands lots of fine-tuning from the user’s side and cannot work on the aesthetic image. So, I propose the solution of the deep learning algorithm. While the dataset and the paper for aesthetic image quality assessment are free to use, we are capable of constructing a mathematical model that can learn the pattern of a dataset, hence, predicting the score of quality. As deep learning is an end-to-end solution, it doesn’t require the setting for the hyperparameter. Therefore, we can reduce most of the fine-tuning parts to make this feature easier to use | ||
== Work report == | |||
25/07/2022: | |||
Conclusion of first model for Aesthetic Image Quality Assessment. | |||
07/08/2022 : | |||
Feature Aesthetic Detection to classify aesthetic image in digiKam using deep learning model. | |||
30/08/2022 : | |||
Improvement of the model for Aesthetic Image Quality Assessment. | |||
== Links to Blogs and other writing == | |||
Main merge request: | |||
* https://invent.kde.org/graphics/digikam/-/merge_requests/181 | |||
My blog for GSoC : | |||
* https://phuockhanhle.github.io/jekyll/update/2022/06/19/gsoc-2022.html | |||
Research repository for aesthetic image quality assessment: | |||
* https://github.com/phuockhanhle/iqs-digikam | |||
Latest revision as of 23:17, 4 September 2022
DigiKam Image Quality Sorter Algorithms Improvement
The main idea of IQS in digiKam is to determine the quality of an image and convert it into a score. This score is based on four factors sabotaging image: blur, noise, exposure, and compression. The current approach helps determine whether images are distorted for one of these reasons. However, the current algorithm also presents some drawbacks: It demands lots of fine-tuning from the user’s side and cannot work on the aesthetic image. So, I propose the solution of the deep learning algorithm. While the dataset and the paper for aesthetic image quality assessment are free to use, we are capable of constructing a mathematical model that can learn the pattern of a dataset, hence, predicting the score of quality. As deep learning is an end-to-end solution, it doesn’t require the setting for the hyperparameter. Therefore, we can reduce most of the fine-tuning parts to make this feature easier to use
Work report
25/07/2022: Conclusion of first model for Aesthetic Image Quality Assessment.
07/08/2022 : Feature Aesthetic Detection to classify aesthetic image in digiKam using deep learning model.
30/08/2022 : Improvement of the model for Aesthetic Image Quality Assessment.
Links to Blogs and other writing
Main merge request:
My blog for GSoC :
Research repository for aesthetic image quality assessment: