62322 Scene Recognition System

The 62322 new scene recognition system uses Yolo v3 object detection model as feature extractor and MLP as scene classifier. Evalution is conducted via k-fold validation method.

SECTION 1: Yolo v3 Object Detection with Tensorflow 2.0

Yolo v3 is an algorithm that uses deep convolutional neural networks to detect objects.

Getting started

Prerequisites

This project is written in Python 3.7 using Tensorflow 2.0 (deep learning), NumPy (numerical computing), Pillow (image processing), OpenCV (computer vision) and seaborn (visualization) packages.

pip install -r requirements.txt

Download Database

Download database (folders images3, images4) from Ong_62322_database folder

Downloading official pretrained weights

For Linux: Let's download official weights pretrained on COCO dataset.

wget -P weights https://pjreddie.com/media/files/yolov3.weights

For Windows: You can download the yolov3 weights by clicking here and adding them to the weights folder.

Using Custom trained weights

Get the info in README_YOLO.md

Save the weights in Tensorflow format

Get the info in README_YOLO.md

Running the model

Get the info in README_YOLO.md

To-Do List

• Finish migration to full TF 2.0 (remove tf.compat.v1)
• Run 62322_fyp2.ipynb for Yolo v3 object detection
• Numbers in console indicates progress

Acknowledgments

Get the info in README_YOLO.md

SECTION 2: MLP as scene classifier

Multilayer perceptron (MLP) is a simple artificial neural network (ANN) model that is capable of carrying out classifications.

To-Do List

• Make sure that ./detections/detection.csv is saved with [80 attributes][1 scene label]
• Run mlp.ipynb for MLP model training and testing

FOLDERS

./data

Images sources

./data/images3

Experiment 1 image input

./data/images4

Experiment 2 image input

GridSearch.ipynb

Grid search cv method that helps find the best parameters among all classifier