Image classification with various deep learning architectures
The goal of the image classification is to correctly predict the subject of an image. For this project, because of the restrictions on resources and time, we worked by using a smaller dataset called Tiny ImageNet, then attempted to train an image classifier using this data. This project implemen...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Theses and Dissertations |
| Language: | English |
| Published: |
2018
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/76031 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | The goal of the image classification is to correctly predict the subject of an image.
For this project, because of the restrictions on resources and time, we worked by
using a smaller dataset called Tiny ImageNet, then attempted to train an image
classifier using this data.
This project implemented some famous Convolutional Neural Networks with various
useful techniques. The deep learning architectures we implemented in this project
include AlexNet, GoogLeNet, ResNet and DenseNet and their several different
versions, the techniques we applied in this project include dropout, data
augmentation, weight decay and snapshot ensembles and cyclic learning rates.
Consequently, we compared the performance of them in image classification to get
the best one with the highest accuracy. |
|---|