Open resource-aided image analytics
Image description is currently a hot research field. Most image description generation networks use only a certain data set to train a neural network, and then use the neural network to describe the input image. However, due to the different distribution of different data sets, the network trained o...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Theses and Dissertations |
| Language: | English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/78706 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Image description is currently a hot research field. Most image description generation networks use only a certain data set to train a neural network, and then use the neural network to describe the input image. However, due to the different distribution of different data sets, the network trained on one training set is difficult to perform well on another data set. The main purpose of this project is to improve the description of the current data set by using additional information on the network. The performance of any local network on different data sets can be improved. At the same time, we have added an adaptive attention mechanism to the LSTM network. Whenever a neural network wants to generate a word, this adaptive mechanism can be used to determine whether or not to consider the characteristics of the image. This mechanism can make the statements generated by the local network more reasonable and conform to the image content. |
|---|