VesselTok-data / README.md
chinmay55's picture
Upload folder using huggingface_hub
1cd222e verified
|
Raw
History Blame Contribute Delete
1.61 kB
# VesselTok Data Release
This archive contains the preprocessed data used to train the **Stage 1 implicit-function autoencoder** in the [VesselTok](https://arxiv.org/abs/2603.18797) codebase.
## Contents
### `all_centerlines/`
Contains centerline graphs (`.vtp`) for all anatomies used in this work:
* ATM'22
* AIIB23
* AeroPath
* CoSTA
* PARSE 2022
* HiPaS
* Pulmonary_AV
* TopCoW
Set this directory as:
```yaml
configs/base_config.yaml
setup.ALL_CENTERLINES_PATH
```
### `sampled_points/`
Contains precomputed occupancy query points at a grid resolution of **512**.
There is one `.npz` file per centerline graph, with filenames matching those in `all_centerlines/`, except that the `.vtp` extension is replaced by `.npz`.
These files constitute the training-time cache used with:
```yaml
im_configs.training.use_precomputed_gt_distance: True
im_configs.training.num_points_per_axis_sampling: 512
```
Place this directory at:
```text
<MC_MESH_PATH>/occupancy/resolution_512/sampled_points
```
where `<MC_MESH_PATH>` corresponds to:
```yaml
configs/base_config.yaml
setup.MC_MESH_PATH
```
## Not Included
The following data are **not** included in this release:
* **Ablation-specific data**
Data used for resolution or occupancy-radius sweeps, sub-/supersampled centerlines, OOD kidney chunks, and related ablation experiments are not included.
* **Other query-point resolutions**
Precomputed query points at resolutions other than **512** are not included.
## Setup Instructions
See the **Data Setup** section of the main repository `README.md` for complete setup instructions.