在科学研究中，从方法论上来讲，都应“先见森林，再见树木”。当前，人工智能学术研究方兴未艾，技术迅猛发展，可谓万木争荣，日新月异。对于AI从业者来说，在广袤的知识森林中，系统梳理脉络，才能更好地把握趋势。为此，我们精选国内外优秀的综述文章，开辟“综述专栏”，敬请关注。

来源：知乎—wanghy

地址：https://zhuanlan.zhihu.com/p/382419598

编辑：人工智能前沿讲习

CVPR 2021全部论文已经放出，网址https://openaccess.thecvf.com/CVPR2021?day=all。特总结自动驾驶相关论文（包含自动驾驶workshop)，文章虽然不多，但是产生了两篇最佳论文候选，都出自Uber ATG。

打包下载：本公众号后台回复【cvpr2021】下载汇总论文

系统

MP3: A Unified Model to Map, Perceive, Predict and Plan（Finalist）

Multi-Modal Fusion Transformer for End-to-End Autonomous Driving

Learning by Watching

仿真

GeoSim: Realistic Video Simulation via Geometry-Aware Composition for Self-Driving （Finalist） 参考论文作者总结：https://zhuanlan.zhihu.com/p/377570852

场景生成

SceneGen: Learning to Generate Realistic Traffic Scenes

Projecting Your View Attentively: Monocular Road Scene Layout Estimation via Cross-view Transformation

AdvSim: Generating Safety-Critical Scenarios for Self-Driving Vehicles

地图

HDMapGen: A Hierarchical Graph Generative Model of High Definition Maps

预测

Shared Cross-Modal Trajectory Prediction for Autonomous Driving

Pedestrian and Ego-vehicle Trajectory Prediction from Monocular Camera

SGCN:Sparse Graph Convolution Network for Pedestrian Trajectory Prediction

Interpretable Social Anchors for Human Trajectory Forecasting in Crowds

Introvert: Human Trajectory Prediction via Conditional 3D Attention

Focus on Local: Detecting Lane Marker from Bottom Up via Key Point

Divide-and-Conquer for Lane-Aware Diverse Trajectory Prediction

LaPred: Lane-Aware Prediction of Multi-Modal Future Trajectories of Dynamic Agents

场景识别

Patch-NetVLAD: Multi-Scale Fusion of Locally-Global Descriptors for Place Recognition

感知

Exploring Intermediate Representation for Monocular Vehicle Pose Estimation

Delving into Localization Errors for Monocular 3D Object Detection

Robust Multimodal Vehicle Detection in Foggy Weather Using Complementary Lidar and Radar Signals

ViP-DeepLab: Learning Visual Perception with Depth-aware Video Panoptic Segmentation

导航

Binary TTC: A Temporal Geofence for Autonomous Navigation

运动估计

Self-Supervised Pillar Motion Learning for Autonomous Driving

Workshop

今年CVPR 也举行了自动驾驶workshop http://cvpr2021.wad.vision/，昨天晚上进行，视频网址（貌似现在视频被关掉了）：

总结录用文章和比赛结果如下：

文章

RAD: Realtime and Accurate 3D Object Detection on Embedded Systems

Latent Space Regularization for Unsupervised Domain Adaptation in Semantic Segmentation

Learning Depth-Guided Convolutions for Monocular 3D Object Detection

Accurate 3D Object Detection using Energy-Based Models

Semi-synthesis: A fast way to produce effective datasets for stereo matching

Multi-task Learning with Attention for End-to-end Autonomous Driving

MVFuseNet: Improving End-to-End Object Detection and Motion Forecasting through Multi-View Fusion of LiDAR Data

LCCNet: LiDAR and Camera Self-Calibration using Cost Volume Network

Soft Cross Entropy Loss and Bottleneck Tri-Cost Volume For Efficient Stereo Depth Prediction

Occlusion Guided Scene Flow Estimation on 3D Point Clouds

Video Class Agnostic Segmentation Benchmark for Autonomous Driving

Rethinking of Radar’s Role: A Camera-Radar Dataset and Systematic Annotator via Coordinate Alignment

Waymo Open Dataset Challenge

MOTION PREDICTION CHALLENGE：

第一名：DenseTNT Waymo Open Dataset Motion Prediction Challenge 1 st Place Solution

第二名：ReCoAt A Deep Learning Framework with Attention Mechanism for Multi-Modal Motion Prediction

INTERACTION PREDICTION CHALLENGE：

第一名：Multi-Modal Interactive Agent Trajectory Prediction Using Heterogeneous Edge-Enhanced Graph Attention Network

荣誉奖：AIR 2 for Interaction Prediction

REAL-TIME 3D CHALLENGE：

第一名：1 st Place Solutions to the Real-time 3D Detection and the Most Efficient Model of the Waymo Open Dataset Challenges 2021

第二名：CenterPoint++ submission to the Waymo Real-time 3D Detection Challenge

第三名：3rd Place Solution of Waymo Open Dataset Challenge 2021 Real-time 3D Detection Track

荣誉奖：Real-time 3D Object Detection using Feature Map Flow

REAL-TIME 2D CHALLENGE：

第一名：1st Place Solution for Waymo Open Dataset Challenge 2021 Real-time 2D Detection

第二名：2nd Place Solution for Waymo Open Dataset Challenge — Real-time 2D Object Detection

第三名：3rd place waymo real-time 2D object detection: yolov5 self-ensemble.

荣誉奖：Object Detection with Camera-wise Training

荣誉奖：Waymo Open Dataset Real-Time 2D Object Detection Challenge

本文仅做学术分享，如有侵权，请联系删文。下载1在「3D视觉工坊」公众号后台回复：3D视觉，即可下载 3D视觉相关资料干货，涉及相机标定、三维重建、立体视觉、SLAM、深度学习、点云后处理、多视图几何等方向。
下载2在「3D视觉工坊」公众号后台回复：3D视觉github资源汇总，即可下载包括结构光、标定源码、缺陷检测源码、深度估计与深度补全源码、点云处理相关源码、立体匹配源码、单目、双目3D检测、基于点云的3D检测、6D姿态估计源码汇总等。
下载3在「3D视觉工坊」公众号后台回复：相机标定，即可下载独家相机标定学习课件与视频网址；后台回复：立体匹配，即可下载独家立体匹配学习课件与视频网址。重磅！3DCVer-学术论文写作投稿 交流群已成立
扫码添加小助手微信，可申请加入3D视觉工坊-学术论文写作与投稿 微信交流群，旨在交流顶会、顶刊、SCI、EI等写作与投稿事宜。同时也可申请加入我们的细分方向交流群，目前主要有3D视觉、CV&深度学习、SLAM、三维重建、点云后处理、自动驾驶、多传感器融合、CV入门、三维测量、VR/AR、3D人脸识别、医疗影像、缺陷检测、行人重识别、目标跟踪、视觉产品落地、视觉竞赛、车牌识别、硬件选型、学术交流、求职交流、ORB-SLAM系列源码交流、深度估计等微信群。
一定要备注：研究方向+学校/公司+昵称，例如：”3D视觉 + 上海交大 + 静静“。请按照格式备注，可快速被通过且邀请进群。原创投稿也请联系。
▲长按加微信群或投稿▲长按关注公众号
3D视觉从入门到精通知识星球：针对3D视觉领域的视频课程（三维重建系列、三维点云系列、结构光系列、手眼标定、相机标定、orb-slam3等视频课程）、知识点汇总、入门进阶学习路线、最新paper分享、疑问解答五个方面进行深耕，更有各类大厂的算法工程人员进行技术指导。与此同时，星球将联合知名企业发布3D视觉相关算法开发岗位以及项目对接信息，打造成集技术与就业为一体的铁杆粉丝聚集区，近2000星球成员为创造更好的AI世界共同进步，知识星球入口：
学习3D视觉核心技术，扫描查看介绍，3天内无条件退款圈里有高质量教程资料、答疑解惑、助你高效解决问题
觉得有用，麻烦给个赞和在看~