Page Comparison

Here you will use the samples previously created to train your model. First you will select the folder in which you have saved your sample in .PLY format, by clicking Image Added and give your new model a name.

You may create a new model from scratch or start off an existing model. When starting from scratch, you will need to select which way to interpret the point cloud by selectionning between geometry, intensity or RGB; and define the "subsampling distance" (the minimum feature size to be recognized) and the "neighborhood radius" (the size of the model analysis sphere).

Once this is done you will distribute your sample between ''Training set'' and ''validation set''. Take note that for best results the number of ''Validation Set'' used should be at least a third of the sample used in the ''Training Set''.

• Maximum number of epochs : Will define the number of iteration the algorithm will repeat to learn from the sample set. Higher the iteration is repeated better the learning will be but longer the process will be ending.
• Number of batch : This option defines how many sample sets added in the project's breakdown will be used per process iteration higher the number more memory will be used. In the case that you run out of memory, reduce this number. 
• Enable GPU acceleration : If checked, the application will run using NVidia GPU memory instead of CPU.
• GPU id : Select GPU card ID if you have more than one in you computer.

Then when all is set, click on Image Added to start the model training. The model will go directly in the install folder and ready to use for your AI classification by selecting it.