The folder is structured such that you can decide which folders to download. More specifically, we provide the data to run neural rendering pipelines in Training_Data. The Training_Data is structured as follow:

Training_Data/
├── Accuracy/
    ├── Logitech/
    ├── iPhone/
        ├── SD/
        ├── PIS3/
        ├── PIS4/
            ├── images/
            ├── sparse/0/
├── Precision/
    ├── Logitech/
    ├── iPhone/
        ├── SD/
        ├── PIS3/
        ├── PIS4/
            ├── Split_01/
            ├── Split_02/
            ├── Split_03/
                ├── images/
                ├── sfm_points/
                ├── sparse/0/

The folder Precision has 3 splits which are taken from a sequence of 150 frames and later split in 3 equal sized sequences. Here there is an additional folder sfm_points which are used by Gaussian splatting methods. These points are the result of the transformation that centers the wound area in the center of the reconstruction bounding box. We noticed this was necessary for neural rendering methods to perform better as a consequence of the dependence to the aabb box.

To evaluate the reconstruction we need to recover the sfm reconstruction coordinates to GT coordinates. In transformations you can find the rotation and translation matrices for each wound. For the precision experiments, in Postprocessing the code mesh_transform.py recovers the centering translation we applied previously.

In the folder Evaluation_Code in a similar structure you can find the code we used to report the metrics in our paper. This will source from the folder Aligned_Meshes the reconstruction meshes which we manually aligned in addition to a PolyLine perimeter to only evaluate the reconstruction in the vicinity of the wound area.

Lastly, we release all the checkpoints of the 3 best methods plus LightGlue_COLMAP used as the SfM method to compute the camera poses and SfM point cloud to run the neural rendering methods. NOTE that we release all the files output by the methods, but they are not necessary to run your method. Only the Training_Data, GT, Evaluation_Code, Postprocessing and transformations should be required to test other methods. These folder together will take < 2GB of storage space.