In the paper we investigate the problem of food classification from images, for the Food-101 dataset extended with 31 additional food classes from Polish cuisine. We adopted transfer learning and firstly measured training times for models such as MobileNet, MobileNetV2, ResNet50, ResNet50V2, ResNet101, ResNet101V2, InceptionV3, InceptionResNetV2, Xception, NasNetMobile and DenseNet, for systems with NVIDIA Tesla V100 (Volta) and NVIDIA GTX 1060 (Pascal) GPUs. We presented inference times corresponding to training the various considered network models, both using a desktop NVIDIA GTX 1060 GPU and an Intel i7-7000 CPU. Subsequently, we investigated the InceptionV3 model in more detail, best in the preliminary tests, regarding the impact of both learning rates (including both various fixed and variable rates) as well as batch sizes on the accuracy of classification, along with training times for various batch sizes. This allowed to identify better learning rate configurations as well as classification performance versus training time.
Authors
Additional information
- DOI
- Digital Object Identifier link open in new tab 10.1007/978-3-031-10539-5_20
- Category
- Aktywność konferencyjna
- Type
- publikacja w wydawnictwie zbiorowym recenzowanym (także w materiałach konferencyjnych)
- Language
- angielski
- Publication year
- 2022
