Hello TBC’ers! If you're ready to unlock the full potential of your point cloud data and go beyond standard classifications, these tips are for you! These insights will guide you in tailoring custom deep learning models to precisely identify objects unique to your projects. By focusing on data preparation, training optimization, and advanced settings, you'll learn how to build more accurate models faster, empowering you to tackle complex classification challenges with confidence.
-
Understand Voxel Size Implications: While you need to select a voxel size, be aware that voxelization involves a balance between classification precision and sample size. A voxel size that is too large can negatively impact precision and reduce detail in the model. Conversely, while a very small voxel size (e.g., 1 cm) retains more detail, it might not always be necessary or optimal for all scenarios. The goal is to select a reasonable voxel size for successful results.
-
Leverage RGB Information: For enhanced training, consider using RGB information from your point cloud data. Depending on the use case, incorporating both Intensity and RGB can lead to improved validation performance, such as with stop signs which are always the same color.
-
Train multiple times with different parameters: Experiment with different voxel sizes and using RGB information when training your model to find the highest accuracy. Additionally, consider your computing power when choosing the “Batches per epoch” as they greatly affect the time it takes to train. To save time, you can stop the training process before the “Max epochs” limit is reached and select the best available model.
-
Utilize Diverse Data Sources: You can train a model for your point clouds using data captured from a drone, mobile mapper, or static scanner. This flexibility allows for model creation regardless of your primary data acquisition method.
-
Optimize Training with Cropping and Regional Focus: To achieve more accurate models faster, consider cropping point clouds. It can also be beneficial to train and apply a classification model to specific point cloud regions.
-
Combine Custom Models with Pre-Trained Classifiers: You are not limited to using only one type of model. You can effectively use your custom model in conjunction with TBC's pre-trained classifiers.
You're all set! Maximize your TBC experience and ensure smooth workflows by utilizing learning & support material most aligned with your needs and personal preferences.
To download the latest version of TBC, you can head to https://trimble.com/downloadTBC
Complete free self-paced workflow-based courses with hands-on guided software experiences on Learn.Trimble: https://learn.trimble.com/pages/422/trimble-business-center-tbc
For additional learning content, visit our TBC Survey and Construction YouTube page:
https://www.youtube.com/user/TBCSurvey/videos