DeepSense Forum

Clear all

[Solved] vision_beam_tracking

New Member
Joined: 2 months ago
Posts: 2
Topic starter  

Dear sir,

I run the code related to " Computer Vision Aided Beam Tracking in A real-world mmWave Deployment", I found that the training and testing datasets are built from Scenario 8 of Deepsense 6G, could you elaborate how did you generate the two datasets

thank you 

New Member
Joined: 2 years ago
Posts: 1

Hi Frotan,

I have updated the code in the GitHub repo of this paper (  

The scripts for generating the datasets can be found in ./data.

- sequence

  - This script separates the scenario 8 data into multiple sequences. Each sequence contains the data where the vehicle passes the base station once.

  - Then, it creates the scenario8_series_train.csv and scenario8_series_test.csv for the training and test datasets. The two datasets contain different (non-overlapping) sequences.


  - This script further processes scenario8_series_train.csv and scenario8_series_train.csv to generate scenario8_series_bbox_train.csv and scenario8_series_bbox_test.csv.

  - These two files are directly used in the training/testing of the neural network model.

  - It splits each sequence into (overlapping) sub-sequences of 13 time steps.

  - For each time step, it prepares the vehicle bounding box (input visual data) and the beam index (ground truth).

  - Each row in scenario8_series_bbox_train/test.csv corresponds to one of these sub-sequences.

To reproduce the data generation, please follow the data preprocessing steps in readme.


Best regards,


frotan liked
New Member
Joined: 2 months ago
Posts: 2
Topic starter  

Dear Shuaifeng,

appreciate your feedback, thank you very much, hopefully I can get in touch with you via email

This post was modified 1 week ago 2 times by frotan


Leave a reply

Author Name

Author Email

Title *

Preview 0 Revisions Saved