Uncertainty quantification with PnP-ULA.

This code shows you how to use sampling algorithms to quantify uncertainty of a reconstruction from incomplete and noisy measurements.

ULA obtains samples by running the following iteration:

\[x_{k+1} = x_k + \alpha \eta \nabla \log p_{\sigma}(x_k) + \eta \nabla \log p(y|x_k) + \sqrt{2 \eta} z_k\]

where \(z_k \sim \mathcal{N}(0, I)\) is a Gaussian random variable, \(\eta\) is the step size and \(\alpha\) is a parameter controlling the regularization.

The PnP-ULA method is described in the paper Laumont et al.[1].

import deepinv as dinv
from deepinv.utils.plotting import plot
import torch
from deepinv.utils import load_example

Load image from the internet

This example uses an image of Messi.

device = dinv.utils.get_freer_gpu() if torch.cuda.is_available() else "cpu"

x = load_example("messi.jpg", img_size=32).to(device)

Selected GPU 0 with 4949.25 MiB free memory

Define forward operator and noise model

This example uses inpainting as the forward operator and Gaussian noise as the noise model.

sigma = 0.1  # noise level
physics = dinv.physics.Inpainting(mask=0.5, img_size=x.shape[1:], device=device)
physics.noise_model = dinv.physics.GaussianNoise(sigma=sigma)

# Set the global random seed from pytorch to ensure reproducibility of the example.
torch.manual_seed(0)

<torch._C.Generator object at 0x7f1228f32e90>

Define the likelihood

Since the noise model is Gaussian, the negative log-likelihood is the L2 loss.

\[-\log p(y|x) \propto \frac{1}{2\sigma^2} \|y-Ax\|^2\]

# load Gaussian Likelihood
likelihood = dinv.optim.data_fidelity.L2(sigma=sigma)

Define the prior

The score a distribution can be approximated using Tweedie’s formula via the deepinv.optim.ScorePrior class.

\[\nabla \log p_{\sigma}(x) \approx \frac{1}{\sigma^2} \left(D(x,\sigma)-x\right)\]

This example uses a pretrained DnCNN model. From a Bayesian point of view, the score plays the role of the gradient of the negative log prior The hyperparameter sigma_denoiser (\(sigma\)) controls the strength of the prior.

In this example, we use a pretrained DnCNN model using the deepinv.loss.FNEJacobianSpectralNorm loss, which makes sure that the denoiser is firmly non-expansive (see Terris et al.[2]), and helps to stabilize the sampling algorithm.

sigma_denoiser = 2 / 255
prior = dinv.optim.ScorePrior(
    denoiser=dinv.models.DnCNN(pretrained="download_lipschitz")
).to(device)

Create the MCMC sampler

Here we use the Unadjusted Langevin Algorithm (ULA) to sample from the posterior defined in deepinv.sampling.ULAIterator. The hyperparameter step_size controls the step size of the MCMC sampler, regularization controls the strength of the prior and iterations controls the number of iterations of the sampler.

regularization = 0.9
step_size = 0.01 * (sigma**2)
iterations = int(5e3) if torch.cuda.is_available() else 10
params = {
    "step_size": step_size,
    "alpha": regularization,
    "sigma": sigma_denoiser,
}
f = dinv.sampling.sampling_builder(
    "ULA",
    prior=prior,
    data_fidelity=likelihood,
    max_iter=iterations,
    params_algo=params,
    thinning=1,
    verbose=True,
)

Generate the measurement

We apply the forward model to generate the noisy measurement.

y = physics(x)

Run sampling algorithm and plot results

The sampling algorithm returns the posterior mean and variance. We compare the posterior mean with a simple linear reconstruction.

mean, var = f.sample(y, physics)

# compute linear inverse
x_lin = physics.A_adjoint(y)

# compute PSNR
print(f"Linear reconstruction PSNR: {dinv.metric.PSNR()(x, x_lin).item():.2f} dB")
print(f"Posterior mean PSNR: {dinv.metric.PSNR()(x, mean).item():.2f} dB")

# plot results
error = (mean - x).abs().sum(dim=1).unsqueeze(1)  # per pixel average abs. error
std = var.sum(dim=1).unsqueeze(1).sqrt()  # per pixel average standard dev.
imgs = [x_lin, x, mean, std / std.flatten().max(), error / error.flatten().max()]
plot(
    imgs,
    titles=["measurement", "ground truth", "post. mean", "post. std", "abs. error"],
)

  0%|          | 0/5000 [00:00<?, ?it/s]
  1%|          | 58/5000 [00:00<00:08, 577.23it/s]
  2%|▎         | 125/5000 [00:00<00:07, 628.88it/s]
  4%|▍         | 192/5000 [00:00<00:07, 642.05it/s]
  5%|▌         | 263/5000 [00:00<00:07, 666.67it/s]
  7%|▋         | 330/5000 [00:00<00:07, 661.81it/s]
  8%|▊         | 409/5000 [00:00<00:06, 704.75it/s]
 10%|▉         | 493/5000 [00:00<00:06, 748.00it/s]
 12%|█▏        | 578/5000 [00:00<00:05, 777.70it/s]
 13%|█▎        | 662/5000 [00:00<00:05, 795.95it/s]
 15%|█▍        | 746/5000 [00:01<00:05, 809.43it/s]
 17%|█▋        | 834/5000 [00:01<00:05, 828.94it/s]
 18%|█▊        | 924/5000 [00:01<00:04, 850.27it/s]
 20%|██        | 1016/5000 [00:01<00:04, 868.52it/s]
 22%|██▏       | 1103/5000 [00:01<00:04, 868.07it/s]
 24%|██▍       | 1191/5000 [00:01<00:04, 869.27it/s]
 26%|██▌       | 1279/5000 [00:01<00:04, 870.78it/s]
 27%|██▋       | 1367/5000 [00:01<00:04, 870.96it/s]
 29%|██▉       | 1455/5000 [00:01<00:04, 856.78it/s]
 31%|███       | 1541/5000 [00:01<00:04, 845.21it/s]
 33%|███▎      | 1626/5000 [00:02<00:04, 839.50it/s]
 34%|███▍      | 1710/5000 [00:02<00:03, 834.59it/s]
 36%|███▌      | 1794/5000 [00:02<00:03, 827.98it/s]
 38%|███▊      | 1880/5000 [00:02<00:03, 835.90it/s]
 39%|███▉      | 1965/5000 [00:02<00:03, 839.85it/s]
 41%|████      | 2051/5000 [00:02<00:03, 843.42it/s]
 43%|████▎     | 2136/5000 [00:02<00:03, 843.23it/s]
 44%|████▍     | 2222/5000 [00:02<00:03, 846.16it/s]
 46%|████▌     | 2308/5000 [00:02<00:03, 848.62it/s]
 48%|████▊     | 2394/5000 [00:02<00:03, 849.61it/s]
 50%|████▉     | 2480/5000 [00:03<00:02, 850.35it/s]
 51%|█████▏    | 2566/5000 [00:03<00:02, 851.54it/s]
 53%|█████▎    | 2652/5000 [00:03<00:02, 851.38it/s]
 55%|█████▍    | 2738/5000 [00:03<00:02, 850.12it/s]
 56%|█████▋    | 2824/5000 [00:03<00:02, 847.74it/s]
 58%|█████▊    | 2910/5000 [00:03<00:02, 849.92it/s]
 60%|█████▉    | 2996/5000 [00:03<00:02, 851.22it/s]
 62%|██████▏   | 3082/5000 [00:03<00:02, 851.03it/s]
 63%|██████▎   | 3168/5000 [00:03<00:02, 851.79it/s]
 65%|██████▌   | 3254/5000 [00:03<00:02, 851.43it/s]
 67%|██████▋   | 3340/5000 [00:04<00:01, 843.52it/s]
 68%|██████▊   | 3425/5000 [00:04<00:01, 827.42it/s]
 70%|███████   | 3508/5000 [00:04<00:01, 819.85it/s]
 72%|███████▏  | 3591/5000 [00:04<00:01, 819.49it/s]
 74%|███████▎  | 3676/5000 [00:04<00:01, 828.10it/s]
 75%|███████▌  | 3762/5000 [00:04<00:01, 835.78it/s]
 77%|███████▋  | 3848/5000 [00:04<00:01, 841.25it/s]
 79%|███████▊  | 3934/5000 [00:04<00:01, 844.38it/s]
 80%|████████  | 4020/5000 [00:04<00:01, 846.73it/s]
 82%|████████▏ | 4106/5000 [00:04<00:01, 848.58it/s]
 84%|████████▍ | 4192/5000 [00:05<00:00, 849.61it/s]
 86%|████████▌ | 4278/5000 [00:05<00:00, 850.35it/s]
 87%|████████▋ | 4364/5000 [00:05<00:00, 851.84it/s]
 89%|████████▉ | 4450/5000 [00:05<00:00, 850.42it/s]
 91%|█████████ | 4536/5000 [00:05<00:00, 851.45it/s]
 92%|█████████▏| 4622/5000 [00:05<00:00, 852.24it/s]
 94%|█████████▍| 4708/5000 [00:05<00:00, 851.99it/s]
 96%|█████████▌| 4794/5000 [00:05<00:00, 851.96it/s]
 98%|█████████▊| 4880/5000 [00:05<00:00, 852.31it/s]
 99%|█████████▉| 4966/5000 [00:05<00:00, 853.25it/s]
100%|██████████| 5000/5000 [00:06<00:00, 830.71it/s]
Iteration 4999, current converge crit. = 1.43E-05, objective = 1.00E-03
Iteration 4999, current converge crit. = 3.42E-04, objective = 1.00E-03
Linear reconstruction PSNR: 8.55 dB
Posterior mean PSNR: 22.31 dB

References:

[1]

Rémi Laumont, Valentin De Bortoli, Andrés Almansa, Julie Delon, Alain Durmus, and Marcelo Pereyra. Bayesian imaging using plug & play priors: when langevin meets tweedie. SIAM Journal on Imaging Sciences, 15(2):701–737, 2022.

[2]

Matthieu Terris, Audrey Repetti, Jean-Christophe Pesquet, and Yves Wiaux. Building firmly nonexpansive convolutional neural networks. In ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 8658–8662. IEEE, 2020.