Projects
[2019] Conids Capture | Platform for 360 Tours Management & Rendering
Conids Capture is a comprehensive platform designed for the creation, management, and rendering of 360 tours, transforming the way virtual spaces are experienced. It offers a suite of tools for crafting immersive tours, including scene setups, smooth transitions, and interactive hotspots. Users can generate 3D models directly from 360 photographs, enhancing the virtual visit experience. The platform also supports the addition of meta-tags to enrich tours with multimedia content. Advanced image processing features like face blurring, auto-alignment, and multi-resolution support ensure high-quality visuals. Interactive floor plans and detailed section measurements further augment the platform's utility, making it an all-encompassing solution for 360 tour management.
[Demo]
[2017] CBIR with Binary Code [Deep Hashing]
This project focuses on creating a Content-Based Image Retrieval (CBIR) system employing deep hashing techniques for efficient and scalable image search. Utilising deep learning, the system transforms images into compact binary codes, facilitating rapid retrieval and accurate ranking among vast datasets. The deep hashing approach ensures that semantically similar images are encoded with closely related binary codes, significantly improving search precision and speed. This advancement addresses the challenges of large-scale image management, enabling users to find relevant images swiftly.
[2017] *Convolutional Feature Masking [Image Segmentation]
Implemented the paper “Convolutional Feature Masking for joint object and stuff segmentation” (CFM).
* a reimplementation from the original paper [Authors: Jifeng Dai, Kaiming He, Jian Sun]
[2016] Faster-RCNN for logo detection [Object Detection]
This project focused on the development of a logo detection system leveraging Faster R-CNN, a state-of-the-art object detection framework. Faster R-CNN combines the strengths of Region Proposal Networks (RPN) with Convolutional Neural Networks (CNN) to efficiently identify and localise logos within images. The system was trained on a dataset of images containing various logos, ensuring robustness and high accuracy across different scenarios. The project's goal was to achieve precise detection in real-time applications, catering to marketing analytics, brand management, and copyright infringement detection. By harnessing Faster R-CNN's capabilities, this initiative provided a valuable tool for automatically analysing visual content for logos.
[2016] Faster-RCNN for face recognition [Face Recognition]
This project involved customizing Faster R-CNN, integrated with Caffe, to create a streamlined face recognition solution. Aimed at providing a simpler alternative to the Torch-based OpenFace, this initiative focused on maintaining comparable performance levels. By adapting Faster R-CNN, known for its efficiency in object detection, for the specific requirements of face recognition, the system offers rapid and accurate identification capabilities. The development process emphasized optimization and fine-tuning to ensure the model's robustness and reliability across various conditions.
[2015] Selective Search for Python
This is a python implementation of the Selective Search
The Selective Search is used as a preprocess of object detection/recognition pipeline. It finds regions likely to contain any objects from an input image regardless of its scale and location, that allows detectors to concentrate only for such 'prospective' regions. Therefore you can configure more computationally efficient detector, or use more rich feature representation and classification methods compared to the conventional exhaustive search scheme.
[2014] Optimisation and Planning Simulator (OPSIM)
OPSIM, the Optimization and Planning Simulator, is a cutting-edge tool designed for the simulation of mobile network technologies, specifically 3G and 4G, facilitating the planning, sizing, and optimization of network resources. It provides telecommunications engineers with a Geographic Information System (GIS)-oriented tool, incorporating geographic data manipulation and mobile network engineering functions (planning, dimensioning, optimization). Developed primarily in Java for its graphical user interface and implementing the MVC (Model-View-Controller) pattern for basic functionalities, OPSIM also harnesses Python for advanced features such as coverage prediction, optimal site positioning, interference analysis, and frequency assignment. The Java component utilizes the ArcObject SDK for Java, while the Python scripts leverage the ArcPy API from ArcGIS, making OPSIM a comprehensive development environment for mobile network engineering.