Open howard19961224 opened 6 years ago
@charlesq34 even I am interested in testing pointnet on a generic point cloud I generated from images using SFM and MVS but the code seems to limit the number of points to be 2048 right? Can we extend it to any pointcoud?
@charlesq34 Even I am very interested in seeing an answer for this question. I am working on a real world outdoor dataset (collected from infrared camera). The point density is about 900,000 points per point cloud. I used Meshlab to downsample it to 1500 points. But I see your algorithm expects the point cloud to be exactly 2048 points. The downsampling process does not explicitly give us an output of exactly 2048 points, but it is always something lesser or higher. I am assuming, we have to do some sort of padding to make up for deficiencies. I am not aware of the details about how we do it ? Could you please guide me to some sample implementation that does something similar ?
https://github.com/hkust-vgd/sese check this
I'm also trying to figure out how to feed my point cloud to the network. At the moment I don't understand exactly what the input for the testing phase is. I can check the code but it's gonna take way longer than looking at a diagram or a quick explanation about inputs and outputs in each stage :) I have a PLY point cloud with position, color, confidence level, intensity, and I need to figure out how to transform this and make it compatible with the batch_inference.py script. Any suggestions?
After two years, has anyone found a solution? I have a Ply dataset from Lidar, and I could not find a way to do segmentation or train my model. I am new to point cloud and lost in the information sea. I just could not find any good solution. Any help? :( :( Thank you!
thank charlesq34 for the code. I get a pointcloud data from realsense camera but how can I do the semantic segmentation with my own data after training. Could you give me any steps or path we can follow?