Recently, two research papers from NavInfo Europe’s AI research team have been accepted at the 2021 Conference on Neural Information Processing Systems (NeurIPS) and will be presented during the virtual workshop of Machine Learning for Autonomous Driving on December 13th. NeurIPS is one of the world’s leading conferences for machine learning and computational neuroscience. Every year in December, top researchers from interdisciplinary academic communities gather to foster the exchange of research advances in Artificial Intelligence and Machine Learning. Furthermore, the conference features professional expositions focussing on machine learning in practice, as well as a series of tutorials and topical workshops that encourage novel approaches to solve critical issues in real-world applications.
As an application of Machine Learning, autonomous driving (AD) has the potential to greatly improve society by reducing road accidents, giving independence to those unable to drive, and even inspiring younger generations with tangible examples of ML-based technology clearly visible on local streets. This workshop on Machine Learning for Autonomous Driving aims to tackle high-impact research problems focussed on perception, gesture recognition, robustness guarantees, real-time constraints, multi-agent planning, and intelligent infrastructure.
Our paper titled “Does Thermal data make the detection systems more reliable?” focuses on solving the challenge of inconsistent performance of the detection networks in different ambient lighting and weather conditions. In this paper, we propose an approach of leveraging data from a different sensor, an Infrared (IR) camera, which is disparate yet complementary to the visual camera data. MultiModal-Collaborative (MMC) framework can integrate both RGB (visual camera) and Thermal (IR camera) modalities to design a comprehensive detection system. In applications where safety is not the utmost criteria, the cost and computational overhead of thermal data might outweigh its benefits. On the other hand, in AD where even a single extra detection can help avoid a disastrous accident, thermal data proves to be more beneficial. We provide a holistic overview by highlighting both benefits and overheads of using a thermal imaging system, thus helping the community in making an informed choice. Watch our video for a detailed explanation of the paper.
Another paper, titled “Self-supervised pretraining for scene change detection” focuses on the challenge of updating HD maps from large ImageNet datasets. We proposed a novel differencing-based self-supervised pretraining method (D-SSCD) for the scene change detection which learns temporal-consistent representations between the pair of images. The D-SSCD uses absolute feature differencing to learn distinctive representations belonging to the changed region directly in an unsupervised way. The results on the VL-CMU-CD and Panoramic change detection datasets demonstrate the superiority of the proposed method over the widely used ImageNet pretraining without using any additional data. Furthermore, our results also demonstrate the robustness of D-SSCD to natural corruptions, out-of-distribution generalization, and its superior performance in limited label scenarios. Watch our video for a detailed explanation of the paper.
About NavInfo Europe AI Research Lab
Our AI Research Lab is an innovation center with a diverse, energetic team of researchers with extensive expertise in machine learning, computer vision, robust AI, and causality. Our strong grip over a broad range of applications enables us to identify AI use cases in various industries and deploy AI technologies for smarter and reliable solutions.
