My current research focuses on developing more efficient and effective techniques for self-supervised representation learning from videos and images, specifically using masked autoencoders. The main challenge of this approach is the high memory requirements and pre-training times, which I am addressing by developing novel algorithms that can reduce the memory footprint of pre-training and accelerate the pre-training process.

Additionally, I am investigating how pre-trained models from one domain can be transferred to another domain to improve performance on downstream tasks, specifically by pre-training on synthetic data and fine-tuning on real data. To address the challenge of transferring learned representations between synthetic and real data, I am developing novel techniques for domain adaptation.

My long-term goal is to develop more effective and efficient self-supervised learning methods that can advance our understanding of visual perception and enable new breakthroughs in fields such as computer vision, robotics, and artificial intelligence more broadly. Ultimately, my research could contribute to the development of powerful tools for visual perception that have applications in areas such as autonomous driving, medical imaging, and surveillance.

Computer Vision

AdaMAE: Adaptive Masking for Efficient Spatiotemporal Learning with Masked Autoencoders (CVPR'23)

Wele Gedara Chaminda Bandara, Naman Patel, Ali Gholami, Mehdi Nikkhah, Motilal Agrawal, Vishal M. Patel

We propose AdaMAE, a novel, adaptive, and end-to-end trainable token sampling strategy for MAEs that takes into account the spatiotemporal properties of all input tokens to sample fewer but informative tokens.

Unite and Conquer: Cross Dataset Multimodal Synthesis using Diffusion Models (CVPR'23)

Nithin Gopalakrishnan Nair, Wele Gedara Chaminda Bandara, Vishal M. Patel

Our method can unite multiple diffusion models trained on multiple sub-tasks and conquer the combined task through our proposed sampling strategy.

Diffuse-Denoise-Count: Accurate Crowd-Counting with Diffusion Models (Under review)

Y. Ranasinghe, N. G. Nair, Wele Gedara Chaminda Bandara, Vishal M. Patel

We propose using conditional diffusion models to predict density maps, as diffusion models are known to model complex distributions well and show high fidelity to training data during crowd-density map generation.

DDPM-CD: Remote Sensing Change Detection using Denoising Diffusion Probabilistic Models (Under review)

Wele Gedara Chaminda Bandara, Nithin Gopalakrishnan Nair, Vishal M. Patel

We propose a novel way to learn the representations from off-the-shelf, unlabeled remote sensing images available by various Earth observation programs through the diffusion process.

HyperTransformer: A Textural and Spectral Feature Fusion Transformer for Pansharpening (CVPR'22)

Wele Gedara Chaminda Bandara, Vishal M. Patel 

We propose a novel transformer network called HyperTransformer for HS pansharpening which achieves significant improvements over SOTA approaches. To the best of our knowledge, we are one of the first to introduce fusion transformer architecture for HS pansharpening.

Orientation-guided Graph Convolutional Network for Bone Surface Segmentation (MICCAI'22)

Aimon Rahman, Wele Gedara Chaminda Bandara, Jeya Maria Jose Valanarasu, Ilker Hacihaliloglu, Vishal M Patel

We propose an orientation-guided graph convolutional network to improve connectivity while segmenting the bone surface.

A Transformer-Based Siamese Network for Change Detection (IGARSS'22)

Wele Gedara Chaminda Bandara, Vishal M. Patel

We presents a transformer-based Siamese network architecture for Change Detection (CD) from a pair of co-registered remote sensing images. The proposed method unifies hierarchically structured transformer encoder with Multi-Layer Perception (MLP) decoder in a Siamese network architecture to efficiently render multi-scale long-range details required for accurate CD.

SPIN Road Mapper: Extracting Roads from Aerial Images via Spatial and Interaction Space Graph Reasoning for Autonomous Driving (ICRA'22)

Wele Gedara Chaminda Bandara, Jeya Maria Jose Valanarasu, Vishal M. Patel

We propose a Spatial and Interaction Space Graph Reasoning (SPIN) module which when plugged into a ConvNet performs reasoning over graphs constructed on spatial and interaction spaces projected from the feature maps. 

Finalists of ICRA 2022 Outstanding Automation paper award !!!

Hyperspectral Pansharpening Based on Improved Deep Image Prior and Residual Reconstruction

Wele Gedara Chaminda Bandara, Jeya Maria Jose Valanarasu, Vishal M. Patel

We proposed a novel approach for HS pansharpening, which mainly consists of three steps: (1) Upsampling the LR-HSI via DIP, (2) Predicting the residual image via over-complete HyperKite, and (3) Obtaining the final fused HSI by summation.

Revisiting Consistency Regularization for Semi-supervised Change Detection in Remote Sensing Images (Under review)

Wele Gedara Chaminda Bandara, Vishal M. Patel

We propose a semi-supervised CD paradigm based on consistency regularization. The proposed approach can effectively leverage the information from freely-available, unlabeled, multi-temporal, remote-sensing images to enhance the CD performance.

Transformer-based SAR Image Despeckling (IGARSS'22) 

Malsha V. Perera, Wele Gedara Chaminda Bandara, Jeya Maria Jose Valanarasu, Vishal M. Patel

We propose a despeckling network comprises of a transformer-based encoder which allows the network to learn global dependencies between different image regions - aiding in better despeckling.

Multispectral and Hyperspectral Image Processing

Validation of multispectral imaging for the detection of selected adulterants in turmeric samples Journal of Food Engineering

Wele Gedara Chaminda Bandara, G  W  K  Prabhath, D  W  S  C  B  Dissanayake, Vijitha R Herath, G  M  R  I  Godaliyadda, M  P  B  Ekanayake, D  Demini, Terrence Madhujith

We proposed a multispectral imaging system for detecting the percentage of the common adulterant; tartrazine colored rice flour found in turmeric powder.

Electrical Power Systems

Coordinated photovoltaic re-phasing: A novel method to maximize renewable energy integration in low voltage networks by mitigating network unbalances Applied Energy

W G Chaminda Bandara, G.M.R.I.Godaliyadda, M.P.B.Ekanayake, J.B.Ekanayake

We proposed a novel method is proposed to mitigate voltage unbalance in LV distribution grids by optimally re-phasing grid-connected rooftop PV systems.

A complete state estimation algorithm for a three-phase four-wire low voltage distribution system with high penetration of solar PV International Journal of Electrical Power & Energy Systems

Wele Gedara Chaminda Bandara, Dilini Almeida, Roshan Indika Godaliyadda, Mervyn Parakrama Ekanayake, Janaka Ekanayake

We proposed a novel approach to estimate the complete state estimation of Low-Voltage Distribution Grids (LVDGs).

A Sensitivity Matrix Approach Using Two-Stage Optimization for Voltage Regulation of LV Networks with High PV Penetration Energies

A.S. Jameel Hassan, Umar Marikkar, G.W. Kasun Prabhath, Aranee Balachandran, W.G. Chaminda Bandara, Parakrama B. Ekanayake, Roshan I. Godaliyadda, Janaka B. Ekanayake

Reactive Power Compensation for Voltage Violations in Distribution Network Conference on Industrial and Information Systems (ICIIS)

Aranee Balachandran, G. W. K. Prabhath, W. G. Chaminda Bandara, G. M. R. I. Godaliyadda, M. P. B. Ekanayake, J. B. Ekanayake