Deep Learning and Convolutional Neural Network

FIND A SOLUTION AT Academic Writers Bay

42028 Deep Learning and Convolutional Neural Network
Assignment 2
“Image Classification & Object Detection”
Kiran Chopra
13462167
Introduction:
The report presents the approaches used to achieve the two objectives of this assessment, namely image classification and object detection. The Fruit 360 dataset was used for the first task, while the Synthetic Fruit dataset was used for the latter.
Image classification was done using VGG16 as the baseline CNN architecture. The following sections discuss in detail reasons behind choosing VGG16, and also provides details about the parameter settings, model summary, customised architecture and results obtained.
Object detection results were obtained using Faster RCNN and SSD models in two separate notebooks. The findings have been discussed in the report.
Dataset
i. Fruit 360
The fruit 360 dataset was used for image classification problem. The dataset contained 131 different classes of fruits, each contained in a separate folder. Further, the Training folder has 67692 images of all these classes, while the Test folder has 22688 images of these 131 classes of fruits.
Here are some of the images from the dataset:
Figure 1: Sample Images of Avocado, Watermelon, Pomegranate from ‘Training’ Folder of Fruit360 dataset
ii. Synthetic Fruit
The Synthetic Fruit dataset was downloaded in a Pascal VOC format from roboflow.com. The dataset has 6000 images of fruits in random backgrounds. The images were divided into Valid and Train folders, with 5000 images in the Train folder, and 1000 in the Valid folder. All images were present in pairs with a jpg and xml files for further processing.
Here are some sample images from the dataset:
Figure 2: Sample images from Train folder of Synthetic Fruit dataset
Proposed CNN Architecture for Image Classification:
Baseline architecture used

READ ALSO...   Selection of your colleagues' responses. - Student Homeworks

[Briefly explain the baseline architecture, layer, activations, optimizer, etc., add an image if possible.]
Customized Architecture
[Briefly explain the customized architecture, and modifications made, add an image if possible.]
Assumption/Intuitions
[Briefly explain the assumption/intuition/basis for the modification(s)/Customization made]
Model Summary
[Provide the model summary of both baseline and customized architectures]
CNN Architecture for Object Detection:
Faster RCNN

READ ALSO...   analyse and evaluate business strategies MG412 Principles of Marketing

[Briefly explain the Faster-RCNN pipeline and the convolution base used (e.g. MobileNet, VGG16, etc.), add an image if possible.]
SDD (Single Shot detector) (or detector of your choice but YOLO)
[Briefly explain the pipeline and the convolution base used, add an image if possible.]
Assumption/intuitions
[Briefly explain the assumption made in the above two-object detection method]
Model Summary
[Provide the model summary of the two object detection methods]
Experimental results and discussion:
Experimental settings:
Image Classification:

[Provide the hyper-parameter settings, data augmentation settings (if applied), transfer learning (if used, provide information on pre-trained model, etc.)]
Object Detection:
[Provide the hyper-parameter settings, data augmentation settings (if applied), transfer learning (if used, provide information on pre-trained model, etc.)]
Experimental results:
Image classification:
Performance on baseline/standard architecture

[Report the performance of the baseline CNN in a tabular format, with accuracy on train, validation and test set. You may add train/validation/test loss as well.]
Performance on customized architecture
[Report the performance of the baseline CNN in a tabular format, with accuracy on train, validation and test set. You may add train/validation/test loss as well.]
Object Detection:
Performance on Faster-RCNN

READ ALSO...   Provide a 2 pages analysis while answering the following question: Week 2 workshop. Prepare this assignment according to the guidelines found in the APA Style Guide. An abstract is required. | StudyDaddy.com - Original paper

[Report the performance of Faster-RCNN in a tabular format, with mAP on train, validation and test set. You may provide other details/analysis as well]
Performance on SDD or any object detector
[Report the performance of SDD/other object detector in a tabular format, with mAP on train, validation and test set. You may provide other details/analysis as well.]
Discussion:
[Provide your understanding on the performance and accuracy obtained. Include some image classification and object detection result images. You may include wrongly classified/detected samples as well.]
Conclusion
[Provide a short paragraph detailing your understanding on the experiments and results.]

Order from Academic Writers Bay
Best Custom Essay Writing Services

QUALITY: 100% ORIGINAL PAPERNO PLAGIARISM – CUSTOM PAPER