RBG-Diff — Pretrained Checkpoints

Pretrained checkpoints for RBG-Diff (Residual-Bootstrapping Generalized Diffusion) sparse-view CT reconstruction, hosted for direct inference / evaluation. Data: HajihajihaJimmy/RBG-Diff-SVCT-data.

Checkpoints

File	Setting	Trained on	View counts
`rbgdiff_sim_ema19.pkl`	Simulation	AAPM-Mayo abdomen	18 / 36 / 72
`rbgdiff_real_ema19.pkl`	Real (Mayo Siemens)	Siemens abdomen + chest	18 / 36 / 72

One checkpoint per setting covers all listed view counts. Both are EMA weights at epoch 19 in RBG-Diff key naming (denoise_fn.recon_* / resid_* / bsrf_*) and strict-load into networks.rbgdiff.RBGDiff.

Download & use

HF_HUB_ENABLE_HF_TRANSFER=1 hf download HajihajihaJimmy/RBG-Diff --local-dir ./checkpoints

Place the files under the RBG-Diff repository's checkpoints/, then run inference with the repo's entry points:

# Simulation (uses checkpoints/rbgdiff_sim_ema19.pkl)
bash test.sh <GPU>

# Real (defaults to checkpoints/rbgdiff_real_ema19.pkl)
python eval_real.py --gpu <GPU> --test_vol_dir <real_test_vol_dir>

Results (AAPM test set, HU window 3000/500)

Views	PSNR (dB)	SSIM (×100)	VIF (×100)
18	40.78	96.65	71.64
36	44.59	98.23	81.06
72	47.98	99.09	88.49

Citation

If you use these checkpoints, please cite the RBG-Diff paper, and cite the dataset they were trained on (required by CC BY 4.0):

@misc{ldct_projection_2020,
  author    = {McCollough, C. and Chen, B. and Holmes III, D. and Duan, X. and
               Yu, Z. and Yu, L. and Leng, S. and Fletcher, J.},
  title     = {Low Dose CT Image and Projection Data (LDCT-and-Projection-data)},
  year      = {2020},
  version   = {7},
  publisher = {The Cancer Imaging Archive},
  doi       = {10.7937/9npb-2637}
}

Data collection was supported by NIBIB grants EB017095 and EB017185.

License

Weights released under the MIT License. They were trained on data licensed CC BY 4.0 — cite the dataset above if you use them. Research use only; not for clinical or diagnostic use.

Downloads last month: -; Downloads are not tracked for this model. How to track