Computer Vision Projects

Project 1: Fruits Classification

Fruit classification plays an important role in many industrial applications including factories, supermarkets and other fields. The importance of fruit classification can also be seen among people with dietary requirements to assist them in selecting the correct categories of fruits.

• Choose the different classes (at least seven classes) from the 131 classes of the Fruits 360 dataset, Report the train, validation and test results.
  Justify your classification results on test dataset by using AUC of ROC curve. Also provide the relevant performance metrics such sensitivity, etc.
• Change the selected classes to Pear (different varieties, Abate, Forelle, Kaiser, Monster, Red, Stone, Williams) or Apples (different varieties: Crimson Snow, Golden, Golden-Red, Granny Smith, Pink Lady, Red, Red Delicious), and report the results providing relevant metrics.
  Justify the difference between the results of this part and the previous.

Project 2: Human Action Detection

Classification of human activities is one of the emerging research areas in the field of computer vision. It can be used in several applications including medical informatics, surveillance, human computer interaction, and task monitoring.

• Choose the different classes (at least seven classes) from the 15 classes of the Human Action Detection dataset, Report the train, validation and test results.
  Justify your classification results on test dataset by using AUC of ROC curve. Also provide the relevant performance metrics such sensitivity, etc.
• Change the selected classes (running, dancing, fighting, cycling, hugging, drinking and eating), and report the results providing relevant metrics.
  Justify the difference between the results of this part and the previous.

Project 3: Echocardiography View Classification

In echocardiography, many canonical view types are possible, each displaying distinct aspects of the heart's complex anatomy. As part of routine clinical care, when images are taken the sonographer is intentionally capturing a specific view, but the annotation of the view type is not applied to the image or recorded in the electronic record. Thus, from raw data alone it is difficult to focus on a specific anatomical view of interest.

• Use TMED-1 to make 3 possible view-type labels available: PLAX, PSAX, or other (meaning something else other than PLAX or PSAX).
  
• Use TMED-2 to find the following classes: PLAX, PSAX, A2C, A4C, or Other
  
• Report the train, validation and test results. Justify your classification results on test dataset by using AUC of ROC curve. Also provide the relevant performance metrics such sensitivity, etc.

Project 4: Forest Cover Segmentation

Forest segmentation finds its application in Land use/Land cover. e.g Forest cover is the amount of land area that is covered by forest. It may be measured as relative or absolute. We can do this by identifying the forest regions and develop a functionality to measure the forest cover from the satellite images itself.

• Segment the forest areas in the satellite images, Report the train, validation and test results. Also provide the relevant performance metrics.
  

Project 5: Full Body Segmentation

Human segmentation often use the outcome of person detection in the video. Segmentation and tracking of the person in the video have significant applications in monitoring and estimating human pose in 2D images and 3D space. The Martial Arts, Dancing and Sports (MADS) dataset, which consists of martial arts actions (Tai-chi and Karate), dancing actions (hip-hop and jazz), and sports actions (basketball, volleyball, football, rugby, tennis and badminton).

• Segment the person in each image, Report the train, validation and test results. Also provide the relevant performance metrics..
  

Project 6: Right/ Left ventricular Segmentation in Echocardiography dataset

The segmentation of Left Ventricle (LV) is currently carried out manually by the experts, and the automation of this process has proved challenging due to the presence of speckle noise and the inherently poor quality of the ultrasound images. The CAMUS dataset, containing 2D apical four-chamber and two-chamber view sequences acquired from 500 patients. The endocardium and epicardium of the left ventricle and left atrium wall are contoured by experts.

• Segment the left ventricle endocardium (LVEndo), the myocardium (epicardium contour more specifically, named LVEpi) and the left atrium (LA), Report the train, validation and test results. Also provide the relevant performance metrics.
  

Project 7: Product Checkout Dataset Object Detection

Over recent years, emerging interest has occurred in integrating computer vision technology into the retail industry. Automatic checkout (ACO) is one of the critical problems in this area which aims to automatically generate the shopping list from the images of the products to purchase. A Large-Scale Retail Product Checkout Dataset(RPC) is the largest one, containing a significant number of product images across various categories.

• Detect object in the test dataset as checkouts. Report the train, validation and test results. Also provide the relevant performance metrics.
  

Project 8: Fruit Images for Object Detection

A fruit dataset for object detection 3 different fruits: Apple, Banana, Orange.

• Detect object in the test dataset(Orange, Banana and apple in bounding box). Report the train, validation and test results. Also provide the relevant performance metrics.
  

Project 9: Early Myocardial Infarction Detection over Multi-view Echocardiography

Myocardial infarction (MI) is the leading cause of mortality in the world that occurs due to a blockage of the coronary arteries feeding the myocardium. An early diagnosis of MI and its localization can mitigate the extent of myocardial damage by facilitating early therapeutic interventions. HMC-QU dataset is the first publicly shared dataset serving myocardial infarction detection on the left ventricle wall. The dataset includes a collection of apical 4-chamber (A4C) and apical 2-chamber (A2C) view 2D-echocardiography recordings.

• Propose a method to detect MI over:
  • Single-view echocardiography by using the information extracted from A4C or A2C views.
  • Multi-view echocardiography by merging the information from A4C or A2C views.


• Report the train, validation and test results. Also provide the relevant performance metrics.

Project 10: Doppler Mitral inflow echocardiographic_ pixel to velocity

The project involves working with Doppler Mitral inflow echocardiographic images. The images show measurements of the blood flow velocity. Currently, AI models can detect and measure necessary parameters on Mitral echocardiographic images, but measurements are acquired in pixel values, while in order to adapt the techniques to real-world scenario, automated measurements should be converted to velocity values.

Usually, there is no way of getting the pixel-to-velocity conversion rate apart from the images themself. Echocardiographic images contain velocity values, as well as the 0-line, so the project requires student to create a deployable model that will process echocardiographic images and will lead to pixel-to-velocity conversion rate. Student is free to use any available computer vision techniques, as long as they can be deployed and incorporated into a larger Deep Learning pipeline.

Student will also be able to have some initial consultation with members of IntSav team, and will work at the AI-lab.

Project 11: Advanced Optimization of Convolutional Neural Networks for Echocardiographic Image Analysis using Automated Hyperparameter Tuning

Introduction:

In the field of medical imaging, particularly echocardiography, Convolutional Neural Networks (CNNs) have shown promising potential to enhance diagnostic processes through accurate image classification and segmentation. However, the optimal performance of CNN models is dependent on the precise tuning of hyperparameters, a task that is traditionally time-consuming and requires extensive experiments. This project proposes the use of automated hyperparameter optimization (HPO) techniques to streamline the optimization process, aiming to improve the efficiency and accuracy of CNN models for echocardiographic analysis. By leveraging advanced HPO methods and tools such as Keras Tuner and Optuna, the project seeks to eliminate the manual tuning bottleneck, fostering advancements in automated medical diagnostics.

Aims:

1. To review and assess the current landscape of CNN architectures and HPO techniques, focusing on their applications and performance in medical imaging, specifically echocardiography.
2. To implement and evaluate automated HPO strategies to enhance CNN models for echocardiographic view classification and image segmentation, comparing their effectiveness against traditional manual tuning methods.

Significance:

This project aims to significantly advance echocardiographic analysis by automating CNN model optimization, enhancing clinical diagnostics and making advanced AI tools more accessible to medical applications and researchers.

Implementation and Datasets:

You can implement automated hyperparameter tuning for the following tasks:

1. Echocardiography View classification using the description and dataset in Project 3.
2. Left ventricular Segmentation in Echocardiography using the description and dataset in Project 6.