search

ghimiredhikura / Complex-YOLOv3

PyTorch implementation of Complex-YOLO paper with YoloV3
GNU General Public License v3.0
190 stars 68 forks source link

Test on my own Lidar data #31

Open chenxyyy opened 4 years ago

chenxyyy commented 4 years ago

Hi, @ghimiredhikura ，thank you for your great work. I met some problem, could you help me? I use this project on my own 16 line Lidar data with the default weights， but I got nothing. I change the conf_thres = 0.0001, then come out some bounding boxes，but they are all wrong.

I want to know if this is normal. What should I do?

zhouzhou33 commented 4 years ago

Are you applying this project to real-time point cloud data?

ghimiredhikura commented 4 years ago

Hi @chenxyyy,

Training with 64 or 32 channel lidar points and testing with 16 channel lidar point won't work. I also tried it, but its not giving anything useful. If you have training data with 16 chennel lidar, it will work. Thanks.

zhouzhou33 commented 4 years ago

@ghimiredhikura Hi, may I ask if this project can be applied to real-time point cloud data?

chenxyyy commented 4 years ago

Hi @zhouzhou33

I have achieved testing on real-time point cloud data!

zhouzhou33 commented 4 years ago

@chenxyyy !!!!!!!!!!May I have your contact information?

chenxyyy commented 4 years ago

Hi @ghimiredhikura

Thank you very much! I decided to reduce the number of KITTI point cloud data to retrain the model. Thank you!

chenxyyy commented 4 years ago

@zhouzhou33 Are you Chinese? My QQ number is 992101327.

abhigoku10 commented 4 years ago

@chenxyyy when you say real time point cloud what does it mean , directly from the sensor ?? are you an Indian thats your contact number

chenxyyy commented 4 years ago

@abhigoku10

Yes, real-time means that get data directly from the sensor.

I am Chinese, not Indian. If you want to communicate with me, you can email me, 992101327@qq.com

nguyenlam185 commented 4 years ago

Hi @ghimiredhikura

Thank you very much! I decided to reduce the number of KITTI point cloud data to retrain the model. Thank you!

Hi @chenxyyy , I am currently confront with the same problem about the number of channels, may I know how to reduce the number of KITTI point cloud as I have no clue about this technique. Thank you

chenxyyy commented 4 years ago

@nguyenlam185 ,In the vertical direction, the range angle is 28.6 °, which is divided into 64 ranges on average. Take a quarter of that as our new data.

nguyenlam185 commented 4 years ago

@nguyenlam185 ,In the vertical direction, the range angle is 28.6 °, which is divided into 64 ranges on average. Take a quarter of that as our new data.

Thank you so much

trns1997 commented 4 years ago

Hi guy, I fall in a similar case as I have data from a RSLIDAR16 lidar. @nguyenlam185 any success on training using 1/4 of the pointcloud data from kitti data sets velodyne data?

nguyenlam185 commented 4 years ago

Hi guy, I fall in a similar case as I have data from a RSLIDAR16 lidar. @nguyenlam185 any success on training using 1/4 of the pointcloud data from kitti data sets velodyne data?

I didn't do it yet, because my sensor turns out that it has the same channels as the Velodyne, it just limits the range of view to only about 60 degree in front of the sensor. So I fed the pointclouds collected from it directly to the AI model and it worked like a charm. Thanks to @ghimiredhikura for your generous of letting people use your code, also thanks to everyone here for sharing ideas and knowledge

Manueljohnson063 commented 3 years ago

@zhouzhou33 Hai , did you got detection using your own lidar data?

dylantzx commented 2 years ago

@nguyenlam185 ,In the vertical direction, the range angle is 28.6 °, which is divided into 64 ranges on average. Take a quarter of that as our new data.

Hello @chenxyyy, I am new to point clouds. May I know how you do this and how was the result?