search

JorgesNofulla / Point-Cloud-Urban-Change-detection

Repository for detecting changes and classifying them in a point cloud dataset
5 stars 0 forks source link

Step-by-step tutorial video #1

Open gertjogiste opened 4 weeks ago

gertjogiste commented 4 weeks ago

Hello!

I would like to test this code using my classified point clouds. Right now, I have a lot of problems running the code, but I understand that this code is still work in progress. Could you please send a step-by-step tutorial video of using the code or update the code itself? That would help me understand what I am doing wrong.

Thanks in advance!

JorgesNofulla commented 4 weeks ago

Hi, thank you for the comment. The code available should be ready for testing. I'll try to make a user tutorial video as soon as possible. The using of the code is very straightforward as you can use the data preparation code available here https://github.com/JorgesNofulla/Point-Cloud-Urban-Change-detection/tree/main/Data_Pre-Processing_Script and then use the Change_Detection_Scripts to train your model with the prepared data. As for trained model, they are still not available and will be updated very soon this year.

Meanwhile i can help you if you are stuck in a specific step or you dont understand something about the repository :)

gertjogiste commented 3 weeks ago

Hello again! I managed to get data pre-processing script to work! There were some minor issues with python package versions. I am still a novice in coding and deep-learning in general, so excuse me if I ask stupid questions.

My main questions right now about pre-processing script: 1) What are best options to change output point cloud labels to specific colours? I see that the matplotlib' Colormap function is used in creating output .las file - should I modify that part (variable "cmap") ? For visualization, I am trying to mainly use point cloud processing software options that are free of charge, so I am trying to get by with CloudCompare.

2) I would like to train the algorithm with several tiles. At the end of the code, there is an option to stack training data. How should I stack them, if the code only creates one training data set per run?

JorgesNofulla commented 2 weeks ago

Haha there are no stupid questions.

  1. For visualization, CloudCompare is a good free option to use for point cloud processing. Yes you can use cmap to pre-set some colours but I recommend using CloudCompare as the software allows you to customize the color settings directly within CloudCompare. You can completely ignore the cmap.

  2. Yeah.. about this. I made a mistake back then by not setting it more automated. A workaround is to run the scrip for your tiles one by one and then stack. Or even better if you can put the training of the model on a loop to work with your saved numpy files instead of needed to stack them. QUICK SOLUTION : So the way you can work with it without changing anything is to simply run the code for every tile until here "print_coord = ahn_train_data" then for the next tile you change it to "print_coord = ahn_train_data2", then for the next "print_coord = ahn_train_data3" and then stack. Also dont foget to do the same for point_labels = targets . you need to change them every time you run the script (you should not restart the script, just simply run it again and again while changing the variable names and of course your data "f" and "f2") I didnt know much myself back then so its a very bad way of doing this :(