2nd Place Solution for Waymo Open Dataset Challenge -- Real-time 2D Object Detection

TL;DR

This paper presents a real-time 2D object detection method for autonomous driving that combines multiple detectors, optimizes inference with TensorRT, and achieves high accuracy with low latency, ranking second in the Waymo Challenge.

Contribution

It introduces a multi-detector ensemble approach optimized for real-time performance and small object detection, leveraging TensorRT for acceleration.

Findings

Achieved 75.00% L1 mAP and 69.72% L2 mAP on Waymo dataset

Operates at 45.8ms/frame on Nvidia Tesla V100

Ranked 2nd in the Waymo Open Dataset Challenge

Abstract

In an autonomous driving system, it is essential to recognize vehicles, pedestrians and cyclists from images. Besides the high accuracy of the prediction, the requirement of real-time running brings new challenges for convolutional network models. In this report, we introduce a real-time method to detect the 2D objects from images. We aggregate several popular one-stage object detectors and train the models of variety input strategies independently, to yield better performance for accurate multi-scale detection of each category, especially for small objects. For model acceleration, we leverage TensorRT to optimize the inference time of our detection pipeline. As shown in the leaderboard, our proposed detection framework ranks the 2nd place with 75.00% L1 mAP and 69.72% L2 mAP in the real-time 2D detection track of the Waymo Open Dataset Challenges, while our framework achieves the latency of 45.8ms/frame on an Nvidia Tesla V100 GPU.