Help with Different Input Pipeline

Hi @josauder @HuguesSib , thanks for the DeepReefMap framework !

I'm planning to test it with deep sea images and already segmented masks with SAM3.

DeepReedMap usability is incredible in running a reconstruction without dependency struggles ^_^ !!!
However, I'm struggling a bit in terms of code documentation/comments and custom design.
It is a bit hard to understand how to modify the project without going through the whole code...
Could I ask you if you have any guidelines on how I can modify the project to do the following:

run the reconstruction with the dataset path in the arguments. For example:

 uv run deepreefmap reconstruct_custom \
  --dataset </path/to/dataset> \
  --camera-profile <camera_name> \
  --mapping <mapper> \
  --out </path/to/output/dir> \
  --viser

I'm planning to organize the dataset as:

/dataset_name
    |--- /images
    |      ... (png/jpg RGB frames) 
    |---/masks    
    |      ... (npy/png/jpg masked frames) 
    |---/configs  
            classes.yaml (similar to yours in /deepreefmap/configs/classes_coralscapes.yaml) 

If you have any project diagram that you may have drafted during development, maybe describing how the different classes/functions and components interact, this would also be super useful for me !

Thanks,
Ale

Start by locating the reconstruction CLI entry point and compare its configuration flow with deepreefmap/configs/classes_coralscapes.yaml. Define how the proposed reconstruct_custom arguments map to the images, masks, and configs dataset layout, then verify that camera-profile, mapping, out, and viser are accepted and produce a reconstruction.