Object Detection with YOLOv8 and YOLOv9 in Machine Vision Using a Manually Collected and Annotated Dataset Coursework

Assessment task details and instructions

In this coursework, you will have the opportunity to apply well- established deep learning techniques learned in this module to analyse datasets that intrigue you. Your task involves conducting a comprehensive analysis, drawing data-driven conclusions, and presenting your findings in a report.

Report Guidelines:

Your report should provide a detailed account of your experimental process, including exploratory data analysis, data preparation,
cleaning, and the algorithms you've chosen. The report should be approximately 3000 words in length (excluding references), following a specified report writing style (worth 25 Marks).

To assist you in structuring your report, a framework is provided. Results must be presented clearly, with in-depth discussions.

Consider using diagrams, as discussed in the lectures, to provide detailed insights into your chosen models and methodologies.

Specific Tasks:

1. Data Collection and Preparation:

Your first task is to collect an image dataset that includes a potential target variable suitable for object detection. While publicly available datasets, like the pig posture dataset and the facial emotional dataset, can be used, I strongly encourage you to explore personally collected datasets. You also have the option to augment existing datasets with images from your collection.

2.  Object Detection :
In this phase, you will apply headless pretrained models, such as YOLOv8 and YOLOv9, using frameworks like PyTorch or Ultralytics. Adapt these models to your specific dataset, and then conduct a comprehensive analysis. Compare and discuss the performance of each model, focusing on aspects like mean average precision and other relevant performance metrics.

3. Comparison of Cloud and Local Computing:

The final task involves implementing your code on both cloud platforms, such as Google Colab, and on local machines, particularly within a virtual environment on Linux. You should provide a clear account of the libraries used in your local environment, including their versions. Discuss the differences in performance between cloud-based and local computing methods, including speed and model training times.

Learning Outcomes

1.   Understand the fundamental principles of deep learning and its applications in object detection.
2.   Grasp the concept of pretrained models and their utility in transfer learning, particularly in object detection tasks.
3.   Gain proficiency in collecting, pre-processing, and cleaning image datasets for object detection purposes.
4.   Learn to select suitable pretrained models for object detection (e.g., YOLOv8 and YOLOv9) and adapt them to specific datasets.
5.   Understand how to evaluate and compare the performance of deep learning models.
6.   Recognise the differences between cloud-based computing (e.g., Google Colab) and local computing within a Linux
virtual environment, and understand how to assess the impact on model training times and performance.