Camera Refinement BIM Helper Functions

This repository contains helper functions for camera refinement with Building Information Modeling (BIM) models with the FACaP (https://github.com/Anna-Ribic/FACaP) pipeline. In order to use these functionalities you need to download the ConSLAM dataset as outlined in their repository.

Helper Functions

Dense Depth from Lidar

• File: projections/dense_depth_from_lidar.py
• Description: Generates a dense depth map from sparse lidar data using interpolation.

Dense Depth with CompletionFormer

• File: projections/CompletionFormer
• Description: Generates a dense depth map from sparse lidar data or another dense depth image using CompletionFormer. Clone the CompletionFormer repo and place iterate.sh and runnet.py in the 'src' folder. Running the iterate.sh script runs the CompletionFormer pipeline for all images in the specified folder path.

Create Synthetic Trajectory

• File: trajectories/model_drift.py
• Description: Generates a .txt file specifying offset trajectory from groundtruth. Groundtruth camera poses for ConSLAM can be found here.

Create Scan for FACaP

• File: projections/create_scan.py

• Description: Creates a Scan folder from a trajectory file. Depth maps can be found here and here.

• File: projections/floorplan_from_json.py

• Description: Generates a floorplan.npy file from a JSON file.

Example Scan

• File: scan_lowres/
• Description: Example Scan for the ConSLAM dataset (https://github.com/mac137/ConSLAM).

Semantic Pointcloud from Depth

• File: projections/semantic_pointcloud_from_depth.py
• Description: Creates a .ply file from semantic masks and depth data.

Evaluation

• File: trajectories/map_metrics.py

• Description: Compute MME, MPV, MOM for pointcloud

• File: trajectories/ate.py

• Description: Compute ATE RMSE and ROT RMSE for estimated and groundtruth trajectory

Visualization

• File: notebooks/Optimization_Vis.ipynb
• Description: Visualize optimization behaviour from FACaP log file