Machine learning for interactive age classification of films
| Reference number | |
| Coordinator | Statens Medieråd |
| Funding from Vinnova | SEK 491 400 |
| Project duration | December 2020 - August 2021 |
| Status | Completed |
| Venture | AI - Competence, ability and application |
| Call | Start your AI-journey for public organizations - autumn 2020 |
Important results from the project
** Denna text är maskinöversatt ** The project´s first objective was to develop an AI-based support for age rating of videos. This has been achieved through the development of AI models that have delivered good results, especially in the case of multimodal models that have been pretrained on other datasets. The project´s second objective was to increase knowledge about AI among the agency´s employees. Lectures and presentations on AI, machine learning and the current project have been carried out. A survey administered before and after the project confirms the increase in knowledge.
Expected long term effects
** Denna text är maskinöversatt ** A dataset consisting of 3600 annotated film clips of 10 sec has been created and made available for download for research purposes. AI models have been created and trained on the dataset. The results are good and several interesting findings have been made, including that the results are improved when image and sound modalities are combined. The results have been described in a scientific article. The project has also led to contacts being made with other actors in the AI field and to new applications being made to further develop the model and dataset.
Approach and implementation
** Denna text är maskinöversatt ** The project has been carried out according to plan and the objectives that were set up have been reached. Several measures were taken to create a dataset of the highest quality possible. The results of the AI model are good and would likely be improved by expanding the data set and make use of more advanced multimodal models. A analysis of the results has been carried out to increase the understanding of what types of content the AI model succeeds or fails to classify correctly. Lessons from that analysis could be used if the model is further developed.