Hi, I have used your framework with great pleasure and I was able to retrain the network with my custom dataset going to obtain excellent performances on my dataset.
Now what I should do is use the network, with my weights, to make inference on pointcloud `` in real time '' that is, without taking the files from a folder and then making inference on all the files together.
What I want to be able to do is:
1: Start the model and keep it active all the time
2: send a pointcloud to the model every certain time (to be read from the disk or even better directly as a 3dpoint vector)
3: receive the labels associated with each 3dpoint directly from python script, without saving them first in a file
4: use in real time the information obtained from semanti segmentation to perform other tasks in my simulator.
From what I understand you need to modify the user.py and infer.py script to do this, could I have some suggestions on how to perform operations 1 and 2 above all?
Question about dataloader
Hi, I have used your framework with great pleasure and I was able to retrain the network with my custom dataset going to obtain excellent performances on my dataset. Now what I should do is use the network, with my weights, to make inference on pointcloud `` in real time '' that is, without taking the files from a folder and then making inference on all the files together. What I want to be able to do is: 1: Start the model and keep it active all the time 2: send a pointcloud to the model every certain time (to be read from the disk or even better directly as a 3dpoint vector) 3: receive the labels associated with each 3dpoint directly from python script, without saving them first in a file 4: use in real time the information obtained from semanti segmentation to perform other tasks in my simulator.
From what I understand you need to modify the user.py and infer.py script to do this, could I have some suggestions on how to perform operations 1 and 2 above all?