nesl / radhar Goto Github PK

Hello Do you have the code to test the training results?

I have trouble getting the classifier's accuracy. Is there is a way for obtaining the accuracies

Hello, your work is impressive and we also want to use the TI mmWave radar to collecte some raw point cloud data. However, while using the ROS package developed by radar-lab, we are not so familiar with that and don't know how to capture the message as well as how to write it to a txt file. I wonder if you are kind to provide some instructions about using the above ROS package to attain the txt file in this repository. Looking forward to your reply, thanks!

voxel function taking so much time due unnecessary looping for coordinate mapping(scaling). the function taking O(N^3) time to scale all the 3D points which slow the over all performance. I have optimise the voxel function to time complexity O(NlogN) but i dont know how to contribute since I am beginner to open source.

Hey author,
Could you please provide the working script to use the TD_CNN_LSTM model for prediction?

which vision of tensorflow and keras should i choose,please? I got errors with python 3.6 , tensorflow 2.0.0 and keras 2.3.1.

I run into out of memory when running code loading data from preprocessed data. The terminal shows the message "KILLED" only. Could anyone here help me resolve the problem?

Hi there! Your work is awe-inspiring, and the codes are pretty clear. I have some questions about the preprocesses. Have you adopted some preprocess methods such as CFAR, clustering algorithms to remove the 3D points caused by static background and just preserve the points from body movement? I also checked the ROS package you mentioned, and there is also no information about this. I am trying to collect some points cloud using TI radar, but I find many noisy points. Could you give me some tips to set the radar parameters to collect such nice data if the dataset is not processed?

I am looking forward to your reply. Thank you very much.

Which tensorflow version should I choose ,please?

Hi, I found a bug in the function voxalize that leads to incorrect results. I checked the value of np.sum(pixel) before return which should be equal to the number of points in the frame, but it is always smaller than it. So I guess some points are not correctly taken into account. I checked the code line by line, and I found line 78 does not work well when the point is close to the maximum boundary(in one of x, y, z dimension). Could you please fix this issue?

Hello,
Your work and the dataset are impressive.
I am going to create the Range-Doppler Map from the provided point could frames (I do not have concrete knowledge about FMCW. It is a native idea).
However, I found that the velocity of the data points (I have checked some of the files) is always 0.0, which is wired based on my native understanding. Could you please tell me something about it?
Looking forward to your reply. Thank you!