MonoGRNet：一种用于单目三维物体检测和定位的几何推理网络

最新推荐文章于 2024-08-20 09:48:56 发布
最新推荐文章于 2024-08-20 09:48:56 发布
阅读量3.8k 收藏 15
点赞数 4
分类专栏： autonomous driving - 自动驾驶 object detection - 目标检测 文章标签： MonoGRNet 一种用于单目三维物体检测和定位的几何推理网络

MonoGRNet

https://github.com/Zengyi-Qin/MonoGRNet

MonoGRNet: A Geometric Reasoning Network for Monocular 3D Object Localization
MonoGRNet：单张图像估测物体三维位置
MonoGRNet：一种用于单目三维物体检测和定位的几何推理网络

geometric [,dʒɪə'metrɪk]：adj. 几何学的，几何学图形的
reason ['riːz(ə)n]：n. 理由，理性，动机 vi. 推论，劝说 vt. 说服，推论，辩论
monocular [mə'nɒkjʊlə]：adj. 单眼的，单眼用的
localization [,ləʊkəlaɪ'zeɪʃən]：n. 定位，局限，地方化
Karlsruhe Institute of Technology and Toyota Technological Institute at Chicago，KITTI
Karlsruhe Institute of Technology (German: Karlsruher Institut für Technologie)，KIT：卡尔斯鲁厄理工学院
Toyota Technological Institute at Chicago，TTIC or TTI-Chicago：芝加哥丰田科技学院
Toyota Technological Institute，TTI：丰田工业大学

1. MonoGRNet: A Geometric Reasoning Network for Monocular 3D Object Localization

在这里插入图片描述

amodal 3D bounding boxes，ABBox-3D：非模态三维边界框
instance depth estimation，IDE：实例深度估计，个体深度估计
instance-level depth estimation，IDE：个体级深度估计
local corner regression：局部角落回归

2. Prerequisites

  • ubuntu 16.04 LTS
  • Python 2.7
  • Tensorflow 1.4.0
prerequisite [priː'rekwɪzɪt]：n. 先决条件 adj. 首要必备的

3. Installation

Clone this repository

git clone https://github.com/Zengyi-Qin/MonoGRNet.git

strong@foreverstrong:~/git_workspace$ git clone https://github.com/Zengyi-Qin/MonoGRNet
Cloning into 'MonoGRNet'...
remote: Enumerating objects: 135, done.
remote: Counting objects: 100% (135/135), done.
remote: Compressing objects: 100% (121/121), done.
remote: Total 135 (delta 38), reused 76 (delta 3), pack-reused 0
Receiving objects: 100% (135/135), 2.67 MiB | 1.11 MiB/s, done.
Resolving deltas: 100% (38/38), done.
Checking connectivity... done.
strong@foreverstrong:~/git_workspace$ 
strong@foreverstrong:~/git_workspace$ cd MonoGRNet/
strong@foreverstrong:~/git_workspace/MonoGRNet$ ls
data     demo.jpg  evals  include  LICENSE    README.md  submodules
decoder  encoder   hypes  inputs   optimizer  setup.py   train.py
strong@foreverstrong:~/git_workspace/MonoGRNet$

Download the KITTI Object Detection Dataset (image, calib and label) and place it into data/KittiBox. The folder should be in the following structure:

data
    KittiBox
        training
            calib
            image_2
            label_2
        train.txt
        val.txt

3D Object Detection Evaluation 2017
http://www.cvlibs.net/datasets/kitti/eval_object.php?obj_benchmark=3d

Download left color images of object data set (12 GB)
http://www.cvlibs.net/download.php?file=data_object_image_2.zip

在这里插入图片描述

spam folder：垃圾邮件文件夹

Download camera calibration matrices of object data set (16 MB)
http://www.cvlibs.net/download.php?file=data_object_calib.zip

在这里插入图片描述

Download training labels of object data set (5 MB)
http://www.cvlibs.net/download.php?file=data_object_label_2.zip

在这里插入图片描述

Download object development kit (1 MB) (including 3D object detection and bird’s eye view evaluation code)
devkit_object.zip

The train-val split train.txt and val.txt are contained in this repository.

在这里插入图片描述

在这里插入图片描述

Compile the Cython module and download the pretrained model:

python setup.py

strong@foreverstrong:~/git_workspace/MonoGRNet$ ll
total 320
drwxrwxr-x 13 strong strong   4096 May 30 14:06 ./
drwxrwxr-x  9 strong strong   4096 May 30 08:15 ../
drwxrwxr-x  2 strong strong   4096 May 30 14:18 3D Object Detection Evaluation 2017/
drwxrwxr-x  3 strong strong   4096 May 30 08:15 data/
drwxrwxr-x  2 strong strong   4096 May 30 08:15 decoder/
-rw-rw-r--  1 strong strong 248497 May 30 08:15 demo.jpg
drwxrwxr-x  2 strong strong   4096 May 30 08:15 encoder/
drwxrwxr-x  2 strong strong   4096 May 30 08:15 evals/
drwxrwxr-x  8 strong strong   4096 May 30 08:15 .git/
drwxrwxr-x  2 strong strong   4096 May 30 08:15 hypes/
drwxrwxr-x  5 strong strong   4096 May 30 08:15 include/
drwxrwxr-x  2 strong strong   4096 May 30 08:15 inputs/
-rw-rw-r--  1 strong strong  11356 May 30 08:15 LICENSE
drwxrwxr-x  2 strong strong   4096 May 30 08:15 optimizer/
-rw-rw-r--  1 strong strong   2180 May 30 08:15 README.md
-rw-rw-r--  1 strong strong    565 May 30 08:15 setup.py
drwxrwxr-x  3 strong strong   4096 May 30 08:15 submodules/
-rw-rw-r--  1 strong strong   2401 May 30 08:15 train.py
strong@foreverstrong:~/git_workspace/MonoGRNet$ 
strong@foreverstrong:~/git_workspace/MonoGRNet$ python setup.py
pip install runcython
/usr/local/lib/python2.7/dist-packages/pip/_vendor/requests/__init__.py:83: RequestsDependencyWarning: Old version of cryptography ([1, 2, 3]) may cause slowdown.
  warnings.warn(warning, RequestsDependencyWarning)
Collecting runcython
  Downloading https://files.pythonhosted.org/packages/e5/8c/14816b8df2b61660b26354545115f9b2fe3ceb032d775a111a08b15ec435/runcython-0.2.5.tar.gz
Requirement already satisfied: Cython>=0.10 in /usr/local/lib/python2.7/dist-packages (from runcython) (0.28.2)
Building wheels for collected packages: runcython
  Running setup.py bdist_wheel for runcython ... done
  Stored in directory: /home/strong/.cache/pip/wheels/77/73/f6/dbf7b8972de1de9c07361c5c2a9a1d2c02022e1168bd2a34b0
Successfully built runcython
Installing collected packages: runcython
Could not install packages due to an EnvironmentError: [Errno 13] Permission denied: '/usr/local/lib/python2.7/dist-packages/runcython-0.2.5.dist-info'
Consider using the `--user` option or check the permissions.

You are using pip version 18.0, however version 19.1.1 is available.
You should consider upgrading via the 'pip install --upgrade pip' command.
Makefile:5: recipe for target 'all-python2' failed
make: *** [all-python2] Error 1
--2019-05-30 15:07:56--  https://cloud.tsinghua.edu.cn/f/d958b1da1c49496fa951/?dl=1
Resolving cloud.tsinghua.edu.cn (cloud.tsinghua.edu.cn)... 101.6.8.7
Connecting to cloud.tsinghua.edu.cn (cloud.tsinghua.edu.cn)|101.6.8.7|:443... connected.
HTTP request sent, awaiting response... 302 Found
Location: https://cloud.tsinghua.edu.cn/seafhttp/files/95f7ae29-bfd0-4b3e-86d5-824bdcfcc20e/data.tar.gz [following]
--2019-05-30 15:07:56--  https://cloud.tsinghua.edu.cn/seafhttp/files/95f7ae29-bfd0-4b3e-86d5-824bdcfcc20e/data.tar.gz
Reusing existing connection to cloud.tsinghua.edu.cn:443.
HTTP request sent, awaiting response... 200 OK
Length: 361164800 (344M) [application/octet-stream]
Saving to: ‘pretrained.tar.gz’

pretrained.tar.gz                         32%[=========================>      ] 112.92M  30.9KB/s    in 2m 28s  

2019-05-30 15:10:25 (780 KB/s) - Connection closed at byte 118406801. Retrying.

--2019-05-30 15:10:26--  (try: 2)  https://cloud.tsinghua.edu.cn/seafhttp/files/95f7ae29-bfd0-4b3e-86d5-824bdcfcc20e/data.tar.gz
Connecting to cloud.tsinghua.edu.cn (cloud.tsinghua.edu.cn)|101.6.8.7|:443... connected.
HTTP request sent, awaiting response... 206 Partial Content
Length: 361164800 (344M), 242757999 (232M) remaining [application/octet-stream]
Saving to: ‘pretrained.tar.gz’

pretrained.tar.gz                        100%[++++++++++++++++++++++++++=====>] 344.43M  2.75MB/s    in 1m 56s  

2019-05-30 15:12:23 (1.99 MB/s) - ‘pretrained.tar.gz’ saved [361164800/361164800]

Pretrained model downloaded.
Setup finished!
strong@foreverstrong:~/git_workspace/MonoGRNet$

setup.py

#!/usr/bin/env python3
# -*- coding:utf-8 -*-

import os

# Cython module
os.system("cd ./include/utils/ && make")

# download the pre-trained model
if not os.path.exists("./data/model_2D.pkl"):
    os.system("wget https://cloud.tsinghua.edu.cn/f/d958b1da1c49496fa951/?dl=1 -O pretrained.tar.gz")
    os.system("tar -xf pretrained.tar.gz")
    os.system("rm pretrained.tar.gz")

print("pretrained model")

# 3D Object Detection Evaluation 2017
if not os.path.exists("./data/KittiBox/training"):
    print("KITTI dataset was not found in ./data/KittiBox/training.")
    exit("ERROR!")

print("SUCCESS!")


python3 setup.py

3.1 Makefile

SHELL := /bin/bash

.PHONY : all-python2 all-python3
all-python2:
	sudo pip install runcython
	makecython++ stitch_wrapper.pyx "" "stitch_rects.cpp ./hungarian/hungarian.cpp"

all-python3:
	sudo pip3 install runcython3
	makecython3++ stitch_wrapper.pyx "" "stitch_rects.cpp ./hungarian/hungarian.cpp"

make all-python2 - make all-python3 - make

strong@foreverstrong:~/git_workspace/MonoGRNet/include/utils$ cat Makefile 
SHELL := /bin/bash

.PHONY : all-python2 all-python3
all-python2:
	sudo pip install runcython
	makecython++ stitch_wrapper.pyx "" "stitch_rects.cpp ./hungarian/hungarian.cpp"

all-python3:
	sudo pip3 install runcython3
	makecython3++ stitch_wrapper.pyx "" "stitch_rects.cpp ./hungarian/hungarian.cpp"

strong@foreverstrong:~/git_workspace/MonoGRNet/include/utils$ 
strong@foreverstrong:~/git_workspace/MonoGRNet/include/utils$ make
sudo pip install runcython
[sudo] password for strong: 
/usr/local/lib/python2.7/dist-packages/pip/_vendor/requests/__init__.py:83: RequestsDependencyWarning: Old version of cryptography ([1, 2, 3]) may cause slowdown.
  warnings.warn(warning, RequestsDependencyWarning)
The directory '/home/strong/.cache/pip/http' or its parent directory is not owned by the current user and the cache has been disabled. Please check the permissions and owner of that directory. If executing pip with sudo, you may want sudo's -H flag.
The directory '/home/strong/.cache/pip' or its parent directory is not owned by the current user and caching wheels has been disabled. check the permissions and owner of that directory. If executing pip with sudo, you may want sudo's -H flag.
Collecting runcython
Requirement already satisfied: Cython>=0.10 in /usr/local/lib/python2.7/dist-packages (from runcython) (0.28.2)
Installing collected packages: runcython
Successfully installed runcython-0.2.5
makecython++ stitch_wrapper.pyx "" "stitch_rects.cpp ./hungarian/hungarian.cpp"
strong@foreverstrong:~/git_workspace/MonoGRNet/include/utils$ 
strong@foreverstrong:~/git_workspace/MonoGRNet/include/utils$ make all-python2
sudo pip install runcython
/usr/local/lib/python2.7/dist-packages/pip/_vendor/requests/__init__.py:83: RequestsDependencyWarning: Old version of cryptography ([1, 2, 3]) may cause slowdown.
  warnings.warn(warning, RequestsDependencyWarning)
The directory '/home/strong/.cache/pip/http' or its parent directory is not owned by the current user and the cache has been disabled. Please check the permissions and owner of that directory. If executing pip with sudo, you may want sudo's -H flag.
The directory '/home/strong/.cache/pip' or its parent directory is not owned by the current user and caching wheels has been disabled. check the permissions and owner of that directory. If executing pip with sudo, you may want sudo's -H flag.
Requirement already satisfied: runcython in /usr/local/lib/python2.7/dist-packages (0.2.5)
Requirement already satisfied: Cython>=0.10 in /usr/local/lib/python2.7/dist-packages (from runcython) (0.28.2)
makecython++ stitch_wrapper.pyx "" "stitch_rects.cpp ./hungarian/hungarian.cpp"
strong@foreverstrong:~/git_workspace/MonoGRNet/include/utils$ 
strong@foreverstrong:~/git_workspace/MonoGRNet/include/utils$ make all-python3
sudo pip3 install runcython3
The directory '/home/strong/.cache/pip/http' or its parent directory is not owned by the current user and the cache has been disabled. Please check the permissions and owner of that directory. If executing pip with sudo, you may want sudo's -H flag.
The directory '/home/strong/.cache/pip' or its parent directory is not owned by the current user and caching wheels has been disabled. check the permissions and owner of that directory. If executing pip with sudo, you may want sudo's -H flag.
Collecting runcython3
  Downloading https://files.pythonhosted.org/packages/51/9f/7bf25cc3daa53472b3afe58742c218138c70a4a0d6af2cdf13d0ff908a4a/runcython3-0.2.7.tar.gz
Requirement already satisfied (use --upgrade to upgrade): Cython>=0.10 in /usr/local/lib/python3.5/dist-packages (from runcython3)
Installing collected packages: runcython3
  Running setup.py install for runcython3 ... done
Successfully installed runcython3-0.2.7
You are using pip version 8.1.1, however version 19.1.1 is available.
You should consider upgrading via the 'pip install --upgrade pip' command.
makecython3++ stitch_wrapper.pyx "" "stitch_rects.cpp ./hungarian/hungarian.cpp"
strong@foreverstrong:~/git_workspace/MonoGRNet/include/utils$ 
strong@foreverstrong:~/git_workspace/MonoGRNet/include/utils$  ll
total 192
drwxrwxr-x 5 strong strong   4096 Jun  3 10:38 ./
drwxrwxr-x 5 strong strong   4096 May 30 08:15 ../
drwxrwxr-x 3 strong strong   4096 May 30 08:15 annolist/
-rw-rw-r-- 1 strong strong   9163 May 30 08:15 data_utils.py
drwxrwxr-x 2 strong strong   4096 May 30 08:15 hungarian/
-rw-rw-r-- 1 strong strong      0 May 30 08:15 __init__.py
drwxrwxr-x 2 strong strong   4096 May 30 08:15 kaffe/
-rw-rw-r-- 1 strong strong    302 Jun  3 10:37 Makefile
-rw-rw-r-- 1 strong strong   1936 May 30 08:15 rect.py
-rw-rw-r-- 1 strong strong   3798 May 30 08:15 stitch_rects.cpp
-rw-rw-r-- 1 strong strong   3177 May 30 08:15 stitch_rects.hpp
-rw-rw-r-- 1 strong strong    117 May 30 08:15 stitch_wrapper.py
-rw-rw-r-- 1 strong strong   3802 May 30 08:15 stitch_wrapper.pyx
-rwxrwxr-x 1 strong strong 118448 Jun  3 10:38 stitch_wrapper.so*
-rw-rw-r-- 1 strong strong   1063 May 30 08:15 TENSORBOX_LICENSE.txt
-rw-rw-r-- 1 strong strong  13396 May 30 08:15 train_utils.py
strong@foreverstrong:~/git_workspace/MonoGRNet/include/utils$ make all-python2
sudo pip install runcython
/usr/local/lib/python2.7/dist-packages/pip/_vendor/requests/__init__.py:83: RequestsDependencyWarning: Old version of cryptography ([1, 2, 3]) may cause slowdown.
  warnings.warn(warning, RequestsDependencyWarning)
The directory '/home/strong/.cache/pip/http' or its parent directory is not owned by the current user and the cache has been disabled. Please check the permissions and owner of that directory. If executing pip with sudo, you may want sudo's -H flag.
The directory '/home/strong/.cache/pip' or its parent directory is not owned by the current user and caching wheels has been disabled. check the permissions and owner of that directory. If executing pip with sudo, you may want sudo's -H flag.
Requirement already satisfied: runcython in /usr/local/lib/python2.7/dist-packages (0.2.5)
Requirement already satisfied: Cython>=0.10 in /usr/local/lib/python2.7/dist-packages (from runcython) (0.28.2)
makecython++ stitch_wrapper.pyx "" "stitch_rects.cpp ./hungarian/hungarian.cpp"
strong@foreverstrong:~/git_workspace/MonoGRNet/include/utils$ ll
total 192
drwxrwxr-x 5 strong strong   4096 Jun  3 10:44 ./
drwxrwxr-x 5 strong strong   4096 May 30 08:15 ../
drwxrwxr-x 3 strong strong   4096 May 30 08:15 annolist/
-rw-rw-r-- 1 strong strong   9163 May 30 08:15 data_utils.py
drwxrwxr-x 2 strong strong   4096 May 30 08:15 hungarian/
-rw-rw-r-- 1 strong strong      0 May 30 08:15 __init__.py
drwxrwxr-x 2 strong strong   4096 May 30 08:15 kaffe/
-rw-rw-r-- 1 strong strong    302 Jun  3 10:37 Makefile
-rw-rw-r-- 1 strong strong   1936 May 30 08:15 rect.py
-rw-rw-r-- 1 strong strong   3798 May 30 08:15 stitch_rects.cpp
-rw-rw-r-- 1 strong strong   3177 May 30 08:15 stitch_rects.hpp
-rw-rw-r-- 1 strong strong    117 May 30 08:15 stitch_wrapper.py
-rw-rw-r-- 1 strong strong   3802 May 30 08:15 stitch_wrapper.pyx
-rwxrwxr-x 1 strong strong 118160 Jun  3 10:44 stitch_wrapper.so*
-rw-rw-r-- 1 strong strong   1063 May 30 08:15 TENSORBOX_LICENSE.txt
-rw-rw-r-- 1 strong strong  13396 May 30 08:15 train_utils.py
strong@foreverstrong:~/git_workspace/MonoGRNet/include/utils$

4. Training and evaluation

Run the training script and specify the GPU to use:

python train.py --gpus 0

The evaluation is done during training. You can adjust the evaluation intervals in hypes/kittiBox.json.

Acknowledgement

Yongqiang Cheng
关注
专栏目录
MonoGRNet: A Geometric Reasoning Network for Monocular 3D Object Localization
abrams90的专栏
08-05 2410
Abstract 利用RGB单张图片在真实3D场景进行目标定位由于图像投影过程几何信息的丢失变得相当困难，而这个工作在场景理解中有相当重要的作用。我们提出了通过观察到的2D投影与未观察到的深度维度进行几何推理，用于单目RGB图像三维定位的MonoGRNet算法。MonoGRNet是一个由四个特定子任务网络组成的单一统一的网络结构，四个子任务分别是2D目标检测，实例深度估计（IDE），3D定位与本...
Python-MonoGRNet一种用于单目三维物体检测和定位的几何推理网络KITTI
08-11
4. **损失函数设计**：为了训练网络，设计了针对单目三维物体检测和定位的特定损失函数，如深度回归损失、边界框回归损失等，以优化网络性能。 5. **数据集与评估标准**：MonoGRNet在KITTI数据集上进行训练和验证。...
匈牙利算法C++实现教程
最新发布
gitblog_01042的博客
08-20 247
匈牙利算法C++实现教程 hungarian-algorithm-cppA C++ wrapper for a hungarian algorithm implementation项目地址:https://gitcode.com/gh_mirrors/hu/hungarian-algorithm-cpp 1. 项目的目录结构及介绍 hungarian-algorithm-cpp/ ├── Hun...
#每天一篇论文 263/365 MonoGRNet: A Geometric Reasoning Network for Monocular 3D ObjectLocalization
流浪机器人的博客
11-13 873
MonoGRNet：一种几何推测单目三维目标定位网络 原文 代码 本文提出一种能够使用单目RGB图像估计3D目标边界盒子的网络，作者使用对3D目标框内的RGB图像进行3D深度估计，代替了之前的pixel leve。并将估计的3D边框进行中心点回归。 ...
MonoGRNet：三维目标定位的几何推理网络
gitblog_00034的博客
05-27 377
MonoGRNet：三维目标定位的几何推理网络 MonoGRNetMonoGRNet: A Geometric Reasoning Network for Monocular 3D Object Detection and Localization | KITTI项目地址:https://gitcode.com/gh_mirrors/mo/MonoGRNet 在当今的计算机视觉领域，3D对象定位...
MonoGRNet Project
既然选择了远方 便只顾风雨兼程 - 永强
06-07 699
MonoGRNet Project 1. Project Interpreter File -> Settings -> Project: MonoGRNet -> Project Interpreter 2. Script parameters 2.1 Run -> Edit Configurations… 2.2 Defaults -> Python 2.3...
基于几何不确定投影网络的单目三维目标检测
oyang} @ sydney.edu.au{ yanglei，yanjunjie} @ sensetime.com@ ustc.edu.cnliuyat@mail.ustc.edu.cn摘要几何投影是单目三维物体检测中一种有效的深度估计方法。它根据高度估计深度，这将数学先验引入深度模型。 ...
单应性损失：一种用于提高单目3D目标检测性能的简洁、通用的可微损失函数
10800单目3D目标检测的单应性损失0顾佳琦1，2，吴伯健1*，范鲁斌1，黄建强1，曹申1，向志宇2，华先胜101 阿里巴巴云计算有限公司，2 浙江大学0摘要0单目3D目标检测是自动驾驶中的一项重要任务。然而，大多数当前...
M3DSSD: 单目3D单级物体探测器的特征对齐和非对称非局部注意
6145M3DSSD：单目3D单级物体探测器Shujie Luo1 Hang Dai3 凌少3，4永定21浙江大学2浙江大学3穆罕默德·本·扎耶德人工智能大学，阿联酋阿布扎比4Inception Institute of Artificial Intelligence，阿布扎比，阿联酋*...
【3D目标检测】Categorical Depth Distribution Network for Monocular 3D Object Detection
可乐大牛的博客
12-15 993
本文是基于单目图像的3D目标检测方法。以往的深度信息学习方法（无论是显示学习还是隐式学习）都存在一些问题，会导致后面边界框定位困难以往基于基于深度的方法将深度估计和目标检测分开训练以往基于网格的方法生成的鸟瞰图表示质量不高为每个像素预测一组深度分布，能够得到高质量的鸟瞰图，使得边界框定位更加准确实现了端到端的训练，也就是单个网络中同时进行深度估计和目标检测没有中间表示的方法,也就是基于2D特征+各种约束实现检测，如2D-3D约束，关键点约束等。没有显示的使用深度信息，定位性能相对较差。
Categorical Depth Distribution Network for 3D Object Detection
02-27
Categorical Depth Distribution Network for Monocular 3D Object Detection翻译
论文阅读笔记：(2021.06 cvpr) Categorical Depth Distribution Network for Monocular 3D Object Detection
chaoqinyou的博客
02-07 830
paper:https://openaccess.thecvf.com/content/CVPR2021/papers/Reading_Categorical_Depth_Distribution_Network_for_Monocular_3D_Object_Detection_CVPR_2021_paper.pdfhttps://openaccess.thecvf.com/content/CVPR2021/papers/Reading_Categorical_Depth_Distribution_Net
论文精读 《CaDNN: Categorical Depth Distribution Network for Monocular 3D Object Detection》
zyw2002的博客
11-17 2383
研究动机：单目3D目标检测难以预测物体的深度单目 3D 目标检测是自动驾驶领域的一个关键问题，因为它与典型的多传感器系统相比配置比较简单。单目 3D 检测的主要挑战在于准确预测物体深度，由于缺乏直接的距离测量，因此必须从物体和场景线索推断出物体深度。许多方法试图直接估计深度以辅助 3D 检测，但性能有限，深度估计不准确。主要工作： 提出CaDDN我们提出的 Categorical Depth Distribution Network(CaDDN)
《Categorical Depth Distribution Network for Monocular 3D Object Detection》论文阅读
dtuyg的博客
04-02 1297
CaDDN介绍摘要 摘要 单目3D目标检测是自动驾驶里面的关键问题，因为和典型的多源传感器系统相比它提供了一个简单的配置。单目3D检测的主要挑战在于准确地预测目标的深度，由于缺乏直接的距离测量，必须从目标和场景线索中推断出目标深度。许多方法尝试直接估计深度来辅助3D检测，但由于深度的不准确性能有限。我们提出的方法，分类深度分布式网络（CaDDN）对每一个像素使用了一个预测的分类深度的分布来将丰富的上下文特征信息投影到三维空间中适当的深度间隔.。 ...
Categorial Depth Distribution Network for Monocular 3D Object Detection翻译
weixin_43889128的博客
12-08 678
与典型的多传感器系统相比，单目三维目标检测是自主车辆的一个关键问题，因为它提供了一种配置简单的解决方案。 单目3D检测的主要挑战在于准确预测对象深度，由于缺乏直接的距离测量，因此必须从对象和场景线索推断对象深度。许多方法试图直接估计深度以辅助3D检测，但由于深度不准确，性能有限。我们提出的解决方案，分类深度分布网络（CaDDN），使用预测的每个像素的分类深度分布，将丰富的上下文特征信息投影到三维空间中适当的深度间隔。然后，我们使用计算效率高的鸟瞰投影和单级检测器来产生最终的输出检测。我们将CaDDN设计为一
MonoGRNet Project - train.py
既然选择了远方 便只顾风雨兼程 - 永强
06-23 766
MonoGRNet Project - train.py 1. 2.
MonoGRNet论文精读
unbekannten的博客
11-24 429
MonoGRNet论文
单目3D目标检测方法CaDDN解读
leo0308的博客
04-07 3786
论文：Categorical Depth Distribution Network for Monocular 3D Object Detection 代码：https://github.com/TRAILab/CaDDN 0 引言 单目3D检测的最大难点在于深度的估计，精确的深度估计是非常困难的， 已有的方法大多估计都不准。 CaDDN 主要的贡献也是在深度估计上， 它的思想是既然精确地深度估计很困难， 那就估计一个大概的深度范围。 具体而言， 就是不直接回归连续的深度值， 而是把深度离散化为一些范围，这
评论 13
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值