SimPB & Perception in Autonomous Driving | DETR, 3D Object Detection, BEV Transformation Explained скачать в хорошем качестве

SimPB & Perception in Autonomous Driving | DETR, 3D Object Detection, BEV Transformation Explained

Welcome to our detailed presentation of Group 5 - Deep Learning Term Paper on SimPB (Simulation-Based Planning Benchmark) — a cutting-edge framework designed to evaluate and benchmark planning systems in autonomous vehicles under diverse and complex scenarios. This presentation is brought to you by Group 5 as part of our term paper project, and is presented by Tharun Chanda, along with contributions from my teammates: Shashi, Shwetika, Palak, Abhigyan, and Hardeep. What You’ll Learn: Before diving into SimPB, we lay down the essential background topics that are crucial for understanding perception and decision-making in autonomous systems: 2D Object Detection: How traditional detectors recognize and classify objects in images using bounding boxes. 3D Object Detection: Extending perception into three dimensions — understanding depth, orientation, and real-world scale using sensors like LiDAR and stereo cameras. Camera to BEV (Bird’s Eye View) Transformation: Techniques to convert front-facing camera data into top-down BEV maps, enabling better scene understanding for planning modules. DETR (DEtection TRansformer): A transformer-based object detector that replaces conventional pipelines with a fully end-to-end architecture. We explore how DETR works, its attention mechanism, set-based prediction, bipartite matching loss, and how it led to future innovations like Deformable DETR. SimPB – Simulation-Based Planning Benchmark: Finally, we present SimPB, which serves as a benchmark for evaluating planning algorithms in simulated driving environments. It helps test how autonomous systems make decisions under uncertainty, handle edge cases, and integrate perception into planning. Why is this important? SimPB bridges the gap between high-level decision-making and low-level sensor inputs, making it a valuable tool in advancing real-world autonomy in vehicles. About the Team: Group 5 This term paper presentation was collaboratively prepared by: Tharun Chanda Shashi Shwetika Palak Abhigyan Hardeep We hope you enjoy this walkthrough of modern perception systems and their integration into planning benchmarks like SimPB. If you found this helpful or informative, don’t forget to like, comment, and subscribe for more presentations on AI, deep learning, and autonomous systems. #deeplearning #ml #machinelearning #cs #iitmandi