Enables voxel model integration for prediction

Improves wireframe extraction and prediction

Adds PointNet classification for edge detection

Adds initial 3D CNN voxel implementation

Improves training setup and parameters

Refactors patch generation and prediction

Adds prediction details to visualization

Adds classification to point cloud patches

Adapts output based on prediction configuration

Adds classification head to FastPointNet

Updates data directory path

Improves fully deep model training and stability.

Moves weight initialization to a utility function

Implements point cloud to wireframe model

Generates point cloud dataset from HoHo25k.

Improves robustness of apex/eave point detection

Updates dataset and model paths

Enhances FastPointNet architecture for improved accuracy.

Renames learning_rate parameter to lr

Improves training efficiency and configuration

Replaces --num-gpus with --gres for GPU allocation.

Improves data loading and augmentation pipeline

Adds vertex class to patch data

Adds slurm batch script and updates dataset path

Improves training process and model initialization

Implements FastPointNet for vertex prediction

Refactors COLMAP point projection for vertex detection

Improves wireframe prediction with edge fitting

Filters segmented point clouds by depth and masks

Improves 3D wireframe prediction and extraction

Adds ade_segmentations/ to .gitignore

Ignores autogenerated build artifacts

Adds color visualization script

Ensures correct data type for numerical entries

Adds inference script and utils

fix

try train

skip first frame

lets use validation set

Move of the utils and visualiztion functions to a separate files

Initial commit

initial commit