- 博客(87)
- 资源 (16)
- 问答 (1)
- 收藏
- 关注
原创 三代自动驾驶系统及主流科技公司自动驾驶技术方案简介
当前自动驾驶系统按技术特点可分为三代,第一代自动驾驶系统以后融合感知、高精地图、IMU+GPS定位、预测、基于优化和搜索的规控系统为代表。第二代自动驾驶系统以BEV框架为基础,感知部分多采用基于transformer的共享主干网络模型的多任务学习框架,地图定位部分开始采用“重感知、轻地图”的路线,规控仍以搜索、优化方法为主。第三代自动驾驶系统以大模型和端到端技术为代表,代表性工作有毫末智行driveGPT,上海人工智能实验室UniAD等,通过大模型+RLHF+人为设计的中间特征学习习人类驾驶技能、驾驶场景理
2023-10-24 21:41:50 891
原创 BEVFusion、BEVFormer方法总结
这样可以避免相机到激光雷达的投影造成的特征密度损失,提高融合的效果。如Figure 1所示,目标是查询自车前方黑色线框处目标信息,方法主要包括两部分:一,基于空间交叉注意力机制对多不同视角摄像头图像特征的查询;BEVFormer原理是通过利用多视角相机图像生成BEV特征并进行3D目标检测的方法,该方法使用了一个基于变换器的编码器和一个基于残差连接的多层感知机编码器,其中通过空间交叉注意力模块实现不同视角摄像头的特征提取,通过时间自注意力模块实现历史特征和当前时刻特征相关性信息的提取。
2023-07-21 18:20:59 1497
原创 Tansformer模型原理与算法流程
对于一个输入序列,自注意力机制的思想是,对于每个位置的输出都由输入序列中所有位置的信息加权求和得到,而权重就是输入特征不同位置之间的相关性。在输入词嵌入(或其他张量)之后,编码器和解码器都会添加位置编码,用于表示序列中词语的位置信息,位置编码通过sin,cos函数实现,可以处理任意长度的序列。在训练过程中,为了防止编码器提前看到目标序列中的后续词语,需要用掩码来遮挡未来的信息,掩码是一个二维矩阵,其中未来信息对应的位置为0,其他位置为1.,最后将所有输出向量拼接起来,得到最终的输出矩阵。
2023-07-21 09:21:12 425
转载 On Choosing a Deep Reinforcement Learning Library
August 5, 2020, Thomas SimoniniAs deep reinforcement learning continues to become one of the most hyped strategies to achieve AGI (aka Artificial General Intelligence) more and more libraries are being developed. But choosing the best for your needs can b
2022-04-15 17:22:59 320
原创 bag-of-word模型的简单python实现
前言如下图所示,bag-of-word即将text即文本表述按单词拆分,组成词典,然后将文本描述按词按是否在词典出现进行编码,组成词语向量。图中’she loves pizza is … are the best’即是词典,下方的编码即某一描述的词向量编码。基于keras Tokenizer实现from keras.preprocessing.text import Tokenizer text = [ 'There was a man', 'The man had a dog',
2021-07-06 15:53:22 692
原创 Matlab数据游标模式实现三维坐标的实时获取与存储 datacursormode
主函数负责生成数据图,并且设置datacursormode的回调函数。实现主要在callback函数里实现,能够实现鼠标单击获取坐标图中的数据。可进一步开发用于实现三维点、点云标注。function mainhFigure= figure;t=0:0.01:2.2;z=104.7*(1-exp(-11*t/30))-26*t;x=33*(1-exp(-11*t/30))*cosd(27);y=33*(1-exp(-11*t/30))*sind(27);scatter3(x,y,z
2021-05-12 17:28:00 2162 3
原创 Similarity of two trees or graph based on networkx graph(zhang-shasha, tree edit distance)
I want to compute the Zhang-Shasha tree-edit distance between 2 trees (zsslibrary). However, my trees are in the form ofnetworkxgraphs (they actually represent DOM html trees). The example in the zssdocumentationshows how to create a tree by hand:f...
2021-04-08 10:06:20 536
原创 Situation: Nan, Inf or huge value in QACC at DOF xx. The simulation is unstable. Time = xxx
[Example] http://www.mujoco.org/forum/index.php?threads/warning-nan-inf-or-huge-value-in-qacc-at-dof-1.3540/Question1:Hello,I am simulating a 7-DoF robotic arm, and wanna assign the position of one joint like following (before 'mj_step(m, d);', in the
2021-03-05 11:26:08 1405 2
原创 latex 排版记录 从期刊投稿模版开始学latex
1,一些基础资源安装参考的这里:最简单傻瓜式的 Latex+Texstudio 安装教学 简单使用:Latex期刊论文模板使用基础教程 已更新最新版(截至2021年2月)lshort文档,可作为工具书参考。相关操作搜索查找即可(这个文档貌似github可以找到)。资源:20210218_z中文latex-lshort.zip 两个入门级的B站教程,不用前期看很多视频,现查现用即可,开始看这两个就够了,有实际操作,有基础讲解,比较适合用来入门。一:从0开始的美赛LaTeX模板使用教学一二:从0
2021-02-18 11:38:58 1180
原创 MPI4py install and test using python and linux
install using the following command:conda install mpi4pytest like this:mpirun -np 6 python mpi_test.pyor according toinstall and usage of mpi4py, the following command also can be used.mpiexec -n 3 python mpi_test.pyif you did not run the pyt..
2021-01-06 21:21:24 660
原创 Create gym environment with your own xml file and training it using her of baseline
In last blog, we created a xml file that can be verified in mujoco simulator. But when i tested it in mujoco py and try to training it using HER of baseline, it showed oneerror, this error means every body must had at least one "site" which would beused ..
2020-12-17 00:19:02 975
原创 Brief introduction to mujoco or gym modeling using xml
To build an environment that can be used for mujoco and gym, there are two main method. First you can build the model by writing a xml file, the second is building CAD model using 3D desigining software such as solidworks,then generate urdf filerelying ...
2020-12-16 21:41:26 214 1
转载 How to create new environments for Gym
foward fromhttps://github.com/openai/gym/blob/master/docs/creating-environments.md Create a new repo called gym-foo, which should also be a PIP package. A good example ishttps://github.com/openai/gym-soccer. It should have at least the follo..
2020-12-14 14:54:01 139
转载 A brief introduction of robotic environment of openai gym-fetch: reach, push, slide, and pick_place.
原文https://medium.com/@Amritpal001/intro-to-robotics-in-openai-fetch-reach-env-automating-robotics-with-reinforcement-learning-part-2b7452f3a5e9In a recent paper by OpenAi —Solving Rubik’s Cube with a Robot Hand — researchers introduce the concept of trai
2020-12-14 14:43:29 1613
原创 ubuntu20.04 mujoco200, mujoco GLEW initalization error: Missing GL version
mujoco GLEW initalization error: Missing GL versionStep 1Run update command to update package repositories and get the latest package information.sudo apt-get update -yStep 2Run the install command with -yflag to quickly install the packages and de
2020-12-04 17:53:10 465
原创 ubuntu 20.04, cudn10.0, cudnn v7.6, support pytorch 1.2 gpu and tensorflow 1.15 GPU, config record.
0, install nvida driver sudo add-apt-repository ppa:graphics-drivers/ppa then, open "software & updates" from "show application" and choose "additional drivers", and choose a fit driver: Using NVIDIA driver metapackage from nvidia-driver-44...
2020-08-21 16:29:42 393
原创 python list 复制改变原list list赋值操作问题
if you use list of Python3, you can not assign new variable a value using "=" directly. For, if the child variable was changed, original variable would be changed.the operation of copy should be used.it was tested as following:>>> a=[1,2,3.
2020-07-30 09:55:33 894
原创 mujoco_py GLEW initalization error: Missing GL version
Run the install command with -yflag to quickly install the packages and dependencies.sudo apt-get install -y libglew-devGot these from : https://zoomadmin.com/HowToInstall/UbuntuPackage/libglew-devand thenStep 3:export LD_PRELOAD=/usr/lib/x86_64-li.
2020-07-28 10:07:17 695
原创 pythorch Conda for cuda10.0 and cudnn7.4
# CUDA 9.2conda install pytorch==1.2.0 torchvision==0.4.0 cudatoolkit=9.2 -c pytorch# CUDA 10.0conda install pytorch==1.2.0 torchvision==0.4.0 cudatoolkit=10.0 -c pytorch# CPU Onlyconda install pytorch==1.2.0 torchvision==0.4.0 cpuonly -c pytorch.
2020-07-18 23:27:02 845
原创 keras rl typeerror: len is not well defined for symbolic tensors. please call x.shape for shape info
Traceback (most recent call last):File "<stdin>", line 1, in <module>File "/opt/python-3.7.6/lib/python3.7/site-packages/rl/agents/dqn.py", line 108, in __init__ if hasattr(model.output, '__len__') and len(model.output) > 1:File "/opt..
2020-07-17 15:45:23 2557
原创 python tkinter 程序简单打包exe
参考博客:https://blog.csdn.net/wangxiaobei2017/article/details/84560223首先需要在Python安装目录下安装pyinstaller。之后将编写好的程序复制到tensorflow 环境Scripts下打开cmd,找到该路径上图路径.../Ananonda3/envs/tensorflow/Scripts/输入 pyinst...
2019-09-17 11:31:57 7159
转载 keras实现常用深度学习模型LeNet,AlexNet,ZFNet,VGGNet,GoogleNet,Resnet
转载出处:https://blog.csdn.net/wmy199216/article/details/71171401#LeNet#coding=utf-8from keras.models import Sequentialfrom keras.layers import Dense,Flattenfrom keras.layers.convolutional import ...
2019-07-16 11:01:42 622 1
原创 Ax+By+C=0 直线一般式拟合 c++/python
资料来源于https://www.codeproject.com/Articles/576228/Line-Fitting-in-Images-Using-Orthogonal-Linear-Reg先行Mark,并附python代码。另如下https://blog.csdn.net/linxue968/article/details/19913661中提供的方法存在问题,已测试(r*v-...
2019-06-10 22:34:53 7616 2
原创 win7结束进程 时,提示“拒绝访问”、“没有此任务的实例运行” taskkill 结束不了进程
原解决方案https://www.cnblogs.com/jieliujas/p/9413064.html需要注意,根据系统位数选择不同pchunter即32位或64位,我开始随便下载的pchunter,打开没有进程显示,后来下载了64位的打开有显示,这时找到结束不了进程,然后 右键--查看进程线程 ,然后将所有的线程给杀光光,再出来到进程选项卡下结束进程。好用的下载地址:http:...
2019-03-27 16:55:05 12207
转载 Keras神经网络转到Android可用的模型
原文地址:https://cloud.tencent.com/developer/article/1370372这是一篇对手册性质的文章,如果你刚好从事AI开发,可以参考这文章来进行模型转换。Keras转TFLite需要三个过程,Keras 转 Tensorflow 固化 Tensorflow 网络到 PB(Protocol Buffer) PB 转 TFLiteKeras 网络...
2019-03-19 22:12:06 753
转载 Keras深度神经网络训练IMDB情感分类的四种方法
原文https://my.oschina.net/gaussik/blog/850642Keras的官方Examples里面展示了四种训练IMDB文本情感分类的方法,借助这4个Python程序,可以对Keras的使用做一定的了解。以下是对各个样例的解析。IMDB 数据集IMDB情感分类数据集是Stanford整理的一套IMDB影评的情感数据,它含有25000个训练样本,25000...
2019-03-19 21:31:33 1306
转载 解决win7 word 2016中不能加载EndNote x7
解决word 2016中不能加载EndNote x7:系统环境是win 7专业版 64位,在安装好office 2016和EndNote x7后,发现在word里并不能加载EndNote x7,这时候需在endnote安装目录里找到如下红框文件:然后在桌面建立临时文件夹将红色框里的文件后拷贝到里面,改其后缀.dat修改为.rar,解压并会得到以下文件:将其拷贝至D:\Progr...
2019-03-17 11:34:41 3518
原创 Google Coral Dev Board
1, Feactures• Edge TPU Module (SOM)o NXP i.MX 8M SOC (Quad-core Cortex-A53, plus Cortex-M4F)o Google Edge TPU ML accelerator coprocessoro Cryptographic coprocessoro Wi-Fi...
2019-03-08 10:03:14 4820 2
原创 pydot` failed to call GraphViz.Please install GraphViz
https://stackoverflow.com/questions/36886711/keras-runtimeerror-failed-to-import-pydot-after-installing-graphviz-and-pydthen i just conda install graphvizconda install pydotplusstart steps inc...
2019-01-11 19:30:32 991
转载 linux little skill 键盘数据符号和实际显示符号不一致 桌面任务栏移动
1, linux键盘数据符号和实际显示符号不一致问题Original Article第一步:sudo dpkg-reconfigure keyboard-configuration第二步:进入图形界面选择微软usb键盘:microsoft natural keyboard por usb/microsoft natural internet keyboard pro第三部:选...
2018-12-13 15:55:26 442
原创 Modify environment variable so that solving Anaconda与ROS共存问题
when i tried to add a publisher of ROS to a tensorflow python file, i got follow error:ImportError: No module named rospkgand i found a follow solution Link问题描述在安装Anaconda后$ cd ~/catkin_ws...
2018-07-21 11:00:36 293
转载 图像的矩,以及利用矩求图像的重心,方向【补充形状方向部分】
数学中的矩原文http://blog.csdn.net/qq826309057/article/details/70039397维基百科关于 Moment (mathematics) 的介绍:In mathematics, a moment is a specific quantitative measure, used in both mechanics and statistics, of t...
2018-03-22 16:06:42 1710
原创 use video_recorder of image_view to recorde video of a image topic
while you need install the image_view package in ros environment.and run roscore and video_recorder. its' param is as following:~filename (string, default: output.avi)Path and name of the ou
2018-01-31 17:32:44 507
转载 OpenCV矩阵运算总结
篇一:OpenCV矩阵运算总结原文地址:http://www.cvvision.cn/273.html著作权归作者所有。商业转载请联系作者获得授权,非商业转载请注明出处。作者:cvvision链接:http://www.cvvision.cn/273.html来源:CV视觉网一、矩阵Mat I,img,I1,I2,dst,A,B;double k,alpha;Scalar
2018-01-31 10:09:32 18036
原创 yolo train:CUDA Error: an illegal memory access was encountered darknet: cuda.c:36:check_error
when i finished clone the darknet, i have test the darknet by using the dog picture.but when i try to training the net, it always show error as following:error description: CUDA Error: an
2017-10-30 15:36:01 9398 2
原创 darknet_ros error shooting: undefined reference to `clReleaseMemObject@OPENCL_1.0'
libkinect2_registration.so: undefined reference to `clReleaseMemObject@OPENCL_1.0'/home/robot/catkin_ws/devel/lib/libkinect2_registration.so: undefined reference to `clEnqueueWriteBuffer@OPENCL_1.0'
2017-10-23 16:17:13 1198
原创 error note: Failed to load nodelet [/kinect2_points_xyzrgb_sd] of type [depth_image_proc/point_cloud
[ERROR] [1508463280.109066485]: Failed to load nodelet [/kinect2_points_xyzrgb_sd] of type [depth_image_proc/point_cloud_xyzrgb] even after refreshing the cache: According to the loaded plugin descrip
2017-10-20 14:47:25 7726 6
原创 install pcl1.7.2
it must be the pcl1.7.2, i test the pcl1.7.1 can not pass the bellow steps tar xvfj pcl-pcl-1.7.2.tar.gzChange the directory to the pcl-pcl-1.7.2 (replace 1.7.2 with the correct version number)
2017-10-18 08:34:15 747
ecml/pkdd 2020 Graph Representation Learning Applications JiezhongQiu
2020-10-19
camshift target tracking
2018-07-21
FPGA nios ucos任务间通信与同步
2016-04-29
在altium designer9 等中使用protell99se的如0805,0603等PCB封装库
2015-05-05
C++数字图像处理从零入门
2014-10-06
Vb将文本框里输入的十进制数据按chr(13)分割,然后转换成&h+16进制字符串形式
2015-05-04
TA创建的收藏夹 TA关注的收藏夹
TA关注的人