Comparison of class activation maps & gradient based class activation (GRAD-CAM) algorithm
This research project aims to investigate the heat map visualization techniques used for classifying images. The focus will be on Class Activation Mapping and Gradient Class Activation Mapping technique. The process includes implementation of the algorithms and proceed to do testing with different i...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Final Year Project |
Language: | English |
Published: |
Nanyang Technological University
2023
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/165964 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
id |
sg-ntu-dr.10356-165964 |
---|---|
record_format |
dspace |
spelling |
sg-ntu-dr.10356-1659642023-04-21T15:37:27Z Comparison of class activation maps & gradient based class activation (GRAD-CAM) algorithm Lim, Cheng Yun Deepu Rajan School of Computer Science and Engineering ASDRajan@ntu.edu.sg Engineering::Computer science and engineering This research project aims to investigate the heat map visualization techniques used for classifying images. The focus will be on Class Activation Mapping and Gradient Class Activation Mapping technique. The process includes implementation of the algorithms and proceed to do testing with different images. The algorithm will be implemented using PyTorch and used on pre-trained models. The dataset used in the experiments were from the ImageNet. CAM uses global average pooling to generate a heatmap, while Grad-CAM uses gradients of the output class score with respect to the feature maps of the last convolutional layer to generate a more localized heatmap. Bachelor of Engineering (Computer Science) 2023-04-17T06:40:13Z 2023-04-17T06:40:13Z 2023 Final Year Project (FYP) Lim, C. Y. (2023). Comparison of class activation maps & gradient based class activation (GRAD-CAM) algorithm. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/165964 https://hdl.handle.net/10356/165964 en SCSE22-0420 application/pdf Nanyang Technological University |
institution |
Nanyang Technological University |
building |
NTU Library |
continent |
Asia |
country |
Singapore Singapore |
content_provider |
NTU Library |
collection |
DR-NTU |
language |
English |
topic |
Engineering::Computer science and engineering |
spellingShingle |
Engineering::Computer science and engineering Lim, Cheng Yun Comparison of class activation maps & gradient based class activation (GRAD-CAM) algorithm |
description |
This research project aims to investigate the heat map visualization techniques used for classifying images. The focus will be on Class Activation Mapping and Gradient Class Activation Mapping technique. The process includes implementation of the algorithms and proceed to do testing with different images. The algorithm will be implemented using PyTorch and used on pre-trained models. The dataset used in the experiments were from the ImageNet. CAM uses global average pooling to generate a heatmap, while Grad-CAM uses gradients of the output class score with respect to the feature maps of the last convolutional layer to generate a more localized heatmap. |
author2 |
Deepu Rajan |
author_facet |
Deepu Rajan Lim, Cheng Yun |
format |
Final Year Project |
author |
Lim, Cheng Yun |
author_sort |
Lim, Cheng Yun |
title |
Comparison of class activation maps & gradient based class activation (GRAD-CAM) algorithm |
title_short |
Comparison of class activation maps & gradient based class activation (GRAD-CAM) algorithm |
title_full |
Comparison of class activation maps & gradient based class activation (GRAD-CAM) algorithm |
title_fullStr |
Comparison of class activation maps & gradient based class activation (GRAD-CAM) algorithm |
title_full_unstemmed |
Comparison of class activation maps & gradient based class activation (GRAD-CAM) algorithm |
title_sort |
comparison of class activation maps & gradient based class activation (grad-cam) algorithm |
publisher |
Nanyang Technological University |
publishDate |
2023 |
url |
https://hdl.handle.net/10356/165964 |
_version_ |
1764208035791634432 |