Haritha Jayasinghe

Recent publications

1. ECCV

   Towards a Density Preserving Objective Function for Learning on Point Sets

   Jayasinghe, Haritha, and Brilakis, Ioannis

   In European Conference on Computer Vision (ECCV) 2024

   Abs Code

   Accurate measurement of the discrepancy between point sets is crucial for point cloud learning tasks. Chamfer distance (CD) is favoured over more effective loss metrics such as Earth Mover’s Distance (EMD) for this purpose due to its computational efficiency. Previous investigations into loss function improvements almost exclusively focus on 3D losses as static metrics, and ignore their dynamic behaviour during training. We show that directly modifying the correspondence criteria can prevent clustering of points during training, and lead to more uniform point distributions. We propose UniformCD, a novel 3D distance metric that prioritises matching the relative local densities of point neighbourhoods when assigning correspondences. Experiments demonstrate that the proposed loss function improves performance on a variety of tasks such as cloud completion, parametric model optimisation, as well as downstream task performance when used for self-supervised learning, achieving SOTA EMD results among point set objective functions. We show that the proposed method exploits local density information to converge towards globally optimum density distributions, narrowing the disparity between CD and EMD.

2. BMVC

   Learnable Geometry and Connectivity Modelling of BIM Objects

   Jayasinghe, Haritha, and Brilakis, Ioannis

   In British Machine Vision Conference (BMVC) 2023

   Abs Code

   Accurate modelling of object geometries and their connectivity is a critical yet often overlooked aspect of the 3D scan to Building Information Model (BIM) pipeline. It is essential for the extraction of high-level structural information of infrastructure. In this paper, we first propose a novel method for parametric modelling of both primitive and non-primitive geometries. Element models are generated from predictions using a differentiable method, enabling both the integration of fitting error into the loss function, as well as further optimisation of predictions using gradient descent. This eliminates the need for custom distance heuristics, allowing for scalability to any object with parametric geometry. We evaluate our method on a novel benchmark and demonstrate that it accurately predicts model parameters despite the presence of occlusions. Moreover, we validate the utility of the extracted parameters by adopting them to infer connectivity between objects in a scan. This is achieved by framing connectivity inference as a link prediction task on a Graph Neural Network (GNN). This method is the first to learn the underlying nature of connectivity relationships within a BIM model, and significantly outperforms current rule-based methods for connectivity inference. Furthermore, we release a new synthetic dataset of industrial facility BIM element scans.

3. CONVR

   Topological Relationship Modelling for Industrial Facility Digitisation Using Graph Neural Networks

   Jayasinghe, Haritha, and Brilakis, Ioannis

   In International Conference on Construction Applications of Virtual Reality 2023
4. IEA/AIE

   Data-Driven Simulation of Ride-Hailing Services Using Imitation and Reinforcement Learning

   Jayasinghe, Haritha, Jayatilaka, Tarindu, Gunawardena, Ravin, and Thayasivam, Uthayasanker

   In International Conference on Industrial, Engineering and Other Applications of Applied Intelligent Systems 2021