Object detection with deep learning in real world scenario
Object detection with deep learning has evolved over the years and continues to be a popular area of research with newer models being released frequently, surpassing its predecessors. However, despite these advancements, there is still a lack of findings on the strengths and weaknesses of different...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2023
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/166134 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Object detection with deep learning has evolved over the years and continues to be a popular area of research with newer models being released frequently, surpassing its predecessors. However, despite these advancements, there is still a lack of findings on the strengths and weaknesses of different models and how these models compare with each other. There is also a lack of research on how different fine-tuning methods affect the performance of these models.
This project aims to understand the different approaches used by 4 different object detection models which resulted in different performances. In addition, different fine-tuning methods were also applied to these models to see if there are any improvements in the performance. By setting a different learning rate for the backbone and head, most of the models were able to perform better. With data augmentation techniques, the models are more robust in detecting objects of different sizes. A software was also developed that allows the user to visualize the output of the object detection result from an image, video, or webcam in the real world. |
|---|