Skip to content

Latest commit

 

History

History
180 lines (153 loc) · 7.41 KB

README.md

File metadata and controls

180 lines (153 loc) · 7.41 KB

Real dataset

gt_demo

Here we describe the data processing steps for FIPS dataset (real X-ray projection, pseudo-GT volume).

You can follow similar ideas to generate your own dataset.

Download raw data

We use three cases from FIPS public dataset .Download raw data and put them into data_generator/real_dataset/FIPS_raw. Arrange data structure as follow.

└── data_generator   
│   └── real_dataset
│   │   └── FIPS_raw
│   │   │   ├── pine
│   │   │   │   ├── 20201111_walnut_.txt
│   │   │   │   ├── 20201111_walnut_0001.tif
│   │   │   │   ├── ...
│   │   │   │   └── 20201111_walnut_0721.tif
│   │   │   ├── seashell
│   │   │   │   ├── 20211124_seashell_.txt
│   │   │   │   ├── 20211124_seashell_0001.tif
│   │   │   │   ├── ...
│   │   │   │   └── 20211124_seashell_0721.tif
│   │   │   └── walnut
│   │   │   │   ├── 20201111_walnut_.txt
│   │   │   │   ├── 20201111_walnut_0001.tif
│   │   │   │   ├── ...
│   │   │   │   └── 20201111_walnut_0721.tif
Case name Dataset source Note
Pine FIPS Download all *_.zip files and *.text configuration file.
Seashell FIPS Download all *_.zip files and *.text configuration file.
Walnut FIPS Download all *_.zip files and *.text configuration file.

Extract X-ray projections from raw data

For FIPS dataset only.

Use HelTomo Matlab Toolbox to extract X-ray projections from raw data. You can install HelTomo Toolbox via Matlab Add-on Explorer or locally.

After installing HelTomo, run fips_process_raw_data.m with Matlab. The extracted projections will be saved in data_generator/real_dataset/FIPS_processed.

└── data_generator   
│   └── real_dataset
│   │   └── FIPS_processed
│   │   │   ├── pine
│   │   │   │   ├── 0001.mat
│   │   │   │   ├── ...
│   │   │   │   └── 0721.mat
│   │   │   ├── seashell
│   │   │   │   ├── 0001.mat
│   │   │   │   ├── ...
│   │   │   │   └── 0721.mat
│   │   │   └── walnut
│   │   │   │   ├── 0001.mat
│   │   │   │   ├── ...
│   │   │   │   └── 0721.mat

Process data and generate pseudo-GT volume.

We preprocess X-ray projections by resizing them to 560x560 and rescale values to around [0, 1]. We also rescale the scene size to fit it into a unit cube. We use FDK to generate pseudo-GT volume from all 721 projections. We then sample N projections uniformly as training set and another 100 projections randomly as test set (for 2D rendering performance).

# 75-view
python data_generator/real_dataset/generate_data_all.py \
  --data data_generator/real_dataset/FIPS_processed \
  --out data/real_dataset \
  --n_train 75 

# 50-view
python data_generator/real_dataset/generate_data_all.py \
  --data data_generator/real_dataset/FIPS_processed \
  --out data/real_dataset \
  --n_train 50 

# 25-view
python data_generator/real_dataset/generate_data_all.py \
  --data data_generator/real_dataset/FIPS_processed \
  --out data/real_dataset \
  --n_train 25

You can visually check the processed volume data with data_generator/check_volume.py, and projection data with data_generator/check_proj.py.

Initialize Gaussians

Use data_generator/initialize_pcd_all.py to generate initialized Gaussian point clouds of all cases. You can also initialize one case with initialize_pcd.py. The generated point cloud will be saved to init/*.npy in each data folder.

# 75-view
python data_generator/initialize_pcd_all.py \
  --data data/real_dataset/cone_ntrain_75_angle_360 \
  --device 0
# 50-view
python data_generator/initialize_pcd_all.py \
  --data data/real_dataset/cone_ntrain_50_angle_360 \
  --device 0
# 25-view
python data_generator/initialize_pcd_all.py \
  --data data/real_dataset/cone_ntrain_25_angle_360 \
  --device 0

# (Optional) Single case example
python initialize_pcd.py --data XXX --output XXX --INIT_ARGS

Finally, data format should be as follows.

├── data
│   ├── real_dataset
│   │   ├── cone_ntrain_75_angle_360
│   │   │   ├── pine
│   │   │   │   ├── proj_train
│   │   │   │   │   ├──  *.npy
│   │   │   │   │   └──  ...
│   │   │   │   ├── proj_test
│   │   │   │   │   ├──  *.npy
│   │   │   │   │   └──  ...
│   │   │   │   ├── init_*.npy
│   │   │   │   ├── meta_data.json
│   │   │   │   └── vol_gt.npy
│   │   │   └── ...
│   │   ├── cone_ntrain_50_angle_360
│   │   └── cone_ntrain_25_angle_360

Reference

We acknowledges authors for providing open source datasets. Please cite them if you use related data.

@misc{FIPS_CT_dataset,
  author       = {The Finnish Inverse Problems Society},
  title        = {X-ray Tomographic Datasets},
  year         = {2024},
  url          = {https://fips.fi/category/open-datasets/x-ray-tomographic-datasets/},
}

@misc{pine_data,
  author       = {Alexander Meaney},
  title        = {{Cone-Beam Computed Tomography Dataset of a Pine 
                   Cone}},
  month        = aug,
  year         = 2022,
  publisher    = {Zenodo},
  version      = {1.1.0},
  doi          = {10.5281/zenodo.6985407},
  url          = {https://doi.org/10.5281/zenodo.6985407}
}

@misc{walnut_data,
  author       = {Alexander Meaney},
  title        = {Cone-Beam Computed Tomography Dataset of a Walnut},
  month        = aug,
  year         = 2022,
  publisher    = {Zenodo},
  version      = {1.1.0},
  doi          = {10.5281/zenodo.6986012},
  url          = {https://doi.org/10.5281/zenodo.6986012}
}

@misc{seashell_data,
  author       = {Kamutta, Emma and
                  Mäkinen, Sofia and
                  Meaney, Alexander},
  title        = {{Cone-Beam Computed Tomography Dataset of a 
                   Seashell}},
  month        = aug,
  year         = 2022,
  publisher    = {Zenodo},
  version      = {1.1.0},
  doi          = {10.5281/zenodo.6983008},
  url          = {https://doi.org/10.5281/zenodo.6983008}
}