Отрывок: It is also better to combine models trained using single-modality ML when we average the predictions of the RGB and Diff models. An ensemble of RGB and Diff models can achieve results that are similar to the results of the RGB and Flow ensemble. Models trained using single-modality ML achieve the best results in the ensemble of three...
Полная запись метаданных
Поле DC | Значение | Язык |
---|---|---|
dc.contributor.author | Komkov, S.A. | - |
dc.contributor.author | Dzabraev, M.D. | - |
dc.contributor.author | Petiushko, A.A. | - |
dc.date.accessioned | 2023-12-29 13:02:58 | - |
dc.date.available | 2023-12-29 13:02:58 | - |
dc.date.issued | 2023-08 | - |
dc.identifier | Dspace\SGAU\20231228\107772 | ru |
dc.identifier.citation | Komkov SA, Dzabraev MD, Petiushko AA. Mutual modality learning for video action classification. Computer Optics 2023; 47(4): 637-649. DOI: 10.18287/2412-6179-CO-1277. | ru |
dc.identifier.uri | https://dx.doi.org/10.18287/2412-6179-CO-1277 | - |
dc.identifier.uri | http://repo.ssau.ru/handle/Zhurnal-Komputernaya-optika/Mutual-modality-learning-for-video-action-classification-107772 | - |
dc.description.abstract | The construction of models for video action classification progresses rapidly. However, the performance of those models can still be easily improved by ensembling with the same models trained on different modalities (e.g. Optical flow). Unfortunately, it is computationally expensive to use several modalities during inference. Recent works examine the ways to integrate advantages of multi-modality into a single RGB-model. Yet, there is still room for improvement. In this paper, we explore various methods to embed the ensemble power into a single model. We show that proper initialization, as well as mutual modality learning, enhances single-modality models. As a result, we achieve state-of-the-art results in the Something-Something-v2 benchmark. | ru |
dc.language.iso | en | ru |
dc.publisher | Самарский национальный исследовательский университет | ru |
dc.relation.ispartofseries | 47;4 | - |
dc.subject | video recognition | ru |
dc.subject | video action classification | ru |
dc.subject | video labeling | ru |
dc.subject | mutual learning | ru |
dc.subject | optical flow | ru |
dc.title | Mutual modality learning for video action classification | ru |
dc.type | Article | ru |
dc.textpart | It is also better to combine models trained using single-modality ML when we average the predictions of the RGB and Diff models. An ensemble of RGB and Diff models can achieve results that are similar to the results of the RGB and Flow ensemble. Models trained using single-modality ML achieve the best results in the ensemble of three... | - |
dc.classindex.scsti | 28.23.37 | - |
Располагается в коллекциях: | Журнал "Компьютерная оптика" |
Файлы этого ресурса:
Файл | Описание | Размер | Формат | |
---|---|---|---|---|
2412-6179_2023_47_4_637-649.pdf | 1.1 MB | Adobe PDF | Просмотреть/Открыть |
Показать базовое описание ресурса
Просмотр статистики
Поделиться:
Все ресурсы в архиве электронных ресурсов защищены авторским правом, все права сохранены.