Skip to content

XiaoyangBi/AIT202-Deep-learning-final-project-image-recognition

Repository files navigation

AIT202-Deep-learning-final-project-image-recognition

Datasets: https://xmueducn-my.sharepoint.com/:f:/g/personal/ait2009357_xmu_edu_my/Euuos8BnXNZMngrXT0PkAEgBVf3AjmTZ1VokYm35tVvCFA?e=XS4NDX

Presentation: https://xmueducn-my.sharepoint.com/:v:/g/personal/ait2009357_xmu_edu_my/EdVy0S4_t1hOkC9IafUk8V8Bg2pU5nDF_Q6UXDhMKFVuEg?e=tGVvMs

PPT slide: https://xmueducn-my.sharepoint.com/:p:/g/personal/ait2009357_xmu_edu_my/EW11Urz-bDBBuzxSo6_0VsIBUZ0z9SPDX6-XSvGX2grSPw?e=q0pHaF

Over the past decade, CNN has started a revolution in the field of image recognition since AlexNet[2] showed its unparalleled performance and potential in 2012. With the rapid development of relating research, CNN-based image recognition networks began to become the mainstream framework in computer vision. With the knowledge of deep learning, we designed and built our model concerning the most advanced neural network researches in recent years. With empirical analysis and accuracy evaluation of previous researches, we chose ResNet as the blueprint of our model, for its innovative creation of shortcut connection. After modification of the depth of the network and normalization, our model finally obtained better performance than the original Resnet when applied to our dataset. In our project, we created our food-relating dataset named FC05, short for Food Category 05, which has 5 categories of common daily foods: rice, drinks, leafy green vegetables, meat, and noodles, with 230 images in each. Though the full potential of our model cannot be realized during the training due to the small size of our dataset, FC05 still proves the usability of our model and may become a complete dataset for food classification based on Chinese diet habits in the future.

Keywords: CNN, Deep Learning, ResNet, FC05 Dataset, Image Classification

About

only for project

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published