A project demonstrating how to use CUDA-PointPillars to deal with cloud points data from lidar.
Apache License 2.0
502
stars
148
forks
source link
After adding PCL rendering on top of the inference using `cuda_pointpillar`, the bounding box results for the same frame of point cloud inference vary and exhibit continuous fluctuations. #96
I have added a visualization rendering part (using PCL) on top of the cuda_pointpillar code. Then, I performed repeated testing on a frame from the KITTI dataset's training set on my PC. The inferred bounding box results vary each time, and there are many clearly incorrect boxes with persistent fluctuations.
My computer's configuration is as follows:
I repeatedly performed inference on the 10th frame of the training set.
The results of both iterations are shown in the images below:
The partial inference logs are as follows:
Could you please tell me the reason for this? Is there an issue with the PCL rendering part?
I entered my own point cloud bin file. For the same point cloud, the results also fluctuated. I found that it occurred in the data preprocessing part. Do you have any solutions?
I have added a visualization rendering part (using PCL) on top of the
I repeatedly performed inference on the 10th frame of the training set.
The results of both iterations are shown in the images below:
The partial inference logs are as follows:
Could you please tell me the reason for this? Is there an issue with the PCL rendering part?
cuda_pointpillar
code. Then, I performed repeated testing on a frame from the KITTI dataset's training set on my PC. The inferred bounding box results vary each time, and there are many clearly incorrect boxes with persistent fluctuations. My computer's configuration is as follows: