Regularization by Denoising (RED) for Super-Resolution.#

We use as plug-in denoiser the Gradient-Step Denoiser (GSPnP) which provides an explicit prior.

Hurault, S., Leclaire, A., & Papadakis, N. “Gradient Step Denoiser for convergent Plug-and-Play” In International Conference on Learning Representations.

import deepinv as dinv
from pathlib import Path
import torch
from torch.utils.data import DataLoader
from deepinv.optim.data_fidelity import L2
from deepinv.optim.prior import RED
from deepinv.optim.optimizers import optim_builder
from deepinv.training import test
from torchvision import transforms
from deepinv.utils.parameters import get_GSPnP_params
from deepinv.utils.demo import load_dataset, load_degradation

Setup paths for data loading and results.#

BASE_DIR = Path(".")
DATA_DIR = BASE_DIR / "measurements"
RESULTS_DIR = BASE_DIR / "results"
DEG_DIR = BASE_DIR / "degradations"

# Set the global random seed from pytorch to ensure
# the reproducibility of the example.
torch.manual_seed(0)
device = dinv.utils.get_freer_gpu() if torch.cuda.is_available() else "cpu"

Load base image datasets and degradation operators.#

In this example, we use the Set3C dataset and a motion blur kernel from Levin et al. (2009).

dataset_name = "set3c"
img_size = 256 if torch.cuda.is_available() else 32
operation = "super-resolution"
val_transform = transforms.Compose(
    [transforms.CenterCrop(img_size), transforms.ToTensor()]
)
dataset = load_dataset(dataset_name, transform=val_transform)

# Generate the degradation operator.
kernel_index = 1
kernel_torch = load_degradation(
    "kernels_12.npy", DEG_DIR / "kernels", index=kernel_index
)
kernel_torch = kernel_torch.unsqueeze(0).unsqueeze(
    0
)  # add batch and channel dimensions

# Use parallel dataloader if using a GPU to fasten training, otherwise, as all computes are on CPU, use synchronous dataloading.
num_workers = 4 if torch.cuda.is_available() else 0

factor = 2  # down-sampling factor
n_channels = 3  # 3 for color images, 1 for gray-scale images
n_images_max = 3  # Maximal number of images to restore from the input dataset
noise_level_img = 0.03  # Gaussian Noise standart deviation for the degradation
p = dinv.physics.Downsampling(
    img_size=(n_channels, img_size, img_size),
    factor=factor,
    filter=kernel_torch,
    device=device,
    noise_model=dinv.physics.GaussianNoise(sigma=noise_level_img),
)

# Generate a dataset in a HDF5 folder in "{dir}/dinv_dataset0.h5'" and load it.
measurement_dir = DATA_DIR / dataset_name / operation
dinv_dataset_path = dinv.datasets.generate_dataset(
    train_dataset=dataset,
    test_dataset=None,
    physics=p,
    device=device,
    save_dir=measurement_dir,
    train_datapoints=n_images_max,
    num_workers=num_workers,
)
dataset = dinv.datasets.HDF5Dataset(path=dinv_dataset_path, train=True)
kernels_12.npy degradation downloaded in degradations/kernels
Dataset has been saved at measurements/set3c/super-resolution/dinv_dataset0.h5

Setup the PnP algorithm. This involves in particular the definition of a custom prior class.#

We use the proximal gradient algorithm to solve the super-resolution problem with GSPnP.

# Parameters of the algorithm to solve the inverse problem
early_stop = True  # Stop algorithm when convergence criteria is reached
crit_conv = "cost"  # Convergence is reached when the difference of cost function between consecutive iterates is
# smaller than thres_conv
thres_conv = 1e-5
backtracking = True
use_bicubic_init = False  # Use bicubic interpolation to initialize the algorithm
batch_size = 1  # batch size for evaluation is necessarily 1 for early stopping and backtracking to work.

# load specific parameters for GSPnP
lamb, sigma_denoiser, stepsize, max_iter = get_GSPnP_params(operation, noise_level_img)

params_algo = {
    "stepsize": stepsize,
    "g_param": sigma_denoiser,
    "lambda": lamb,
}

# Select the data fidelity term
data_fidelity = L2()


# The GSPnP prior corresponds to a RED prior with an explicit `g`.
# We thus write a class that inherits from RED for this custom prior.
class GSPnP(RED):
    r"""
    Gradient-Step Denoiser prior.
    """

    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self.explicit_prior = True

    def forward(self, x, *args, **kwargs):
        r"""
        Computes the prior :math:`g(x)`.

        :param torch.tensor x: Variable :math:`x` at which the prior is computed.
        :return: (torch.tensor) prior :math:`g(x)`.
        """
        return self.denoiser.potential(x, *args, **kwargs)


method = "GSPnP"
denoiser_name = "gsdrunet"
# Specify the Denoising prior
prior = GSPnP(denoiser=dinv.models.GSDRUNet(pretrained="download").to(device))


# we want to output the intermediate PGD update to finish with a denoising step.
def custom_output(X):
    return X["est"][1]


# instantiate the algorithm class to solve the IP problem.
model = optim_builder(
    iteration="PGD",
    prior=prior,
    g_first=True,
    data_fidelity=data_fidelity,
    params_algo=params_algo,
    early_stop=early_stop,
    max_iter=max_iter,
    crit_conv=crit_conv,
    thres_conv=thres_conv,
    backtracking=backtracking,
    get_output=custom_output,
    verbose=False,
)

# Set the model to evaluation mode. We do not require training here.
model.eval()
Downloading: "https://huggingface.co/deepinv/gradientstep/resolve/main/GSDRUNet_torch.ckpt?download=true" to /home/runner/.cache/torch/hub/checkpoints/GSDRUNet_torch.ckpt

  0%|          | 0.00/64.9M [00:00<?, ?B/s]
  1%|          | 640k/64.9M [00:00<00:12, 5.61MB/s]
  2%|▏         | 1.25M/64.9M [00:00<00:12, 5.24MB/s]
  3%|▎         | 1.88M/64.9M [00:00<00:13, 5.05MB/s]
  4%|▎         | 2.38M/64.9M [00:00<00:13, 4.94MB/s]
  4%|▍         | 2.88M/64.9M [00:00<00:12, 5.01MB/s]
  5%|▌         | 3.38M/64.9M [00:00<00:12, 5.05MB/s]
  6%|▌         | 4.00M/64.9M [00:00<00:12, 5.16MB/s]
  7%|▋         | 4.62M/64.9M [00:00<00:11, 5.38MB/s]
  8%|▊         | 5.25M/64.9M [00:01<00:11, 5.45MB/s]
  9%|▉         | 5.88M/64.9M [00:01<00:11, 5.19MB/s]
 10%|▉         | 6.38M/64.9M [00:01<00:12, 5.11MB/s]
 11%|█         | 7.00M/64.9M [00:01<00:11, 5.28MB/s]
 12%|█▏        | 7.62M/64.9M [00:01<00:11, 5.38MB/s]
 13%|█▎        | 8.25M/64.9M [00:01<00:11, 5.28MB/s]
 14%|█▎        | 8.88M/64.9M [00:01<00:11, 5.34MB/s]
 15%|█▍        | 9.50M/64.9M [00:01<00:11, 5.13MB/s]
 15%|█▌        | 10.0M/64.9M [00:02<00:11, 5.01MB/s]
 16%|█▌        | 10.5M/64.9M [00:02<00:11, 5.01MB/s]
 17%|█▋        | 11.0M/64.9M [00:02<00:11, 4.97MB/s]
 18%|█▊        | 11.5M/64.9M [00:02<00:11, 4.96MB/s]
 18%|█▊        | 12.0M/64.9M [00:02<00:11, 5.03MB/s]
 19%|█▉        | 12.5M/64.9M [00:02<00:11, 4.97MB/s]
 20%|██        | 13.1M/64.9M [00:02<00:10, 5.12MB/s]
 21%|██        | 13.6M/64.9M [00:02<00:10, 4.95MB/s]
 22%|██▏       | 14.1M/64.9M [00:02<00:10, 4.97MB/s]
 23%|██▎       | 14.6M/64.9M [00:03<00:10, 4.93MB/s]
 23%|██▎       | 15.1M/64.9M [00:03<00:10, 4.97MB/s]
 24%|██▍       | 15.8M/64.9M [00:03<00:09, 5.19MB/s]
 25%|██▌       | 16.4M/64.9M [00:03<00:09, 5.24MB/s]
 26%|██▌       | 17.0M/64.9M [00:03<00:09, 5.29MB/s]
 27%|██▋       | 17.6M/64.9M [00:03<00:09, 5.44MB/s]
 28%|██▊       | 18.2M/64.9M [00:03<00:09, 5.25MB/s]
 29%|██▉       | 18.9M/64.9M [00:03<00:09, 5.26MB/s]
 30%|███       | 19.5M/64.9M [00:03<00:08, 5.40MB/s]
 31%|███       | 20.1M/64.9M [00:04<00:08, 5.30MB/s]
 32%|███▏      | 20.8M/64.9M [00:04<00:08, 5.46MB/s]
 33%|███▎      | 21.4M/64.9M [00:04<00:08, 5.52MB/s]
 34%|███▍      | 22.0M/64.9M [00:04<00:08, 5.60MB/s]
 35%|███▍      | 22.6M/64.9M [00:04<00:07, 5.60MB/s]
 36%|███▌      | 23.2M/64.9M [00:04<00:07, 5.56MB/s]
 37%|███▋      | 23.9M/64.9M [00:04<00:07, 5.64MB/s]
 38%|███▊      | 24.5M/64.9M [00:04<00:07, 5.58MB/s]
 39%|███▊      | 25.1M/64.9M [00:05<00:07, 5.61MB/s]
 40%|███▉      | 25.8M/64.9M [00:05<00:07, 5.46MB/s]
 41%|████      | 26.4M/64.9M [00:05<00:07, 5.46MB/s]
 42%|████▏     | 27.0M/64.9M [00:05<00:07, 5.60MB/s]
 43%|████▎     | 27.6M/64.9M [00:05<00:06, 5.79MB/s]
 44%|████▎     | 28.2M/64.9M [00:05<00:06, 5.77MB/s]
 45%|████▍     | 29.0M/64.9M [00:05<00:06, 6.08MB/s]
 46%|████▌     | 29.8M/64.9M [00:05<00:05, 6.29MB/s]
 47%|████▋     | 30.4M/64.9M [00:05<00:05, 6.23MB/s]
 48%|████▊     | 31.0M/64.9M [00:06<00:05, 6.11MB/s]
 49%|████▊     | 31.6M/64.9M [00:06<00:05, 6.02MB/s]
 50%|████▉     | 32.2M/64.9M [00:06<00:05, 6.11MB/s]
 51%|█████     | 32.9M/64.9M [00:06<00:05, 6.23MB/s]
 52%|█████▏    | 33.6M/64.9M [00:06<00:05, 6.41MB/s]
 53%|█████▎    | 34.2M/64.9M [00:06<00:04, 6.45MB/s]
 54%|█████▍    | 35.0M/64.9M [00:06<00:04, 6.47MB/s]
 55%|█████▍    | 35.6M/64.9M [00:06<00:04, 6.20MB/s]
 56%|█████▌    | 36.4M/64.9M [00:06<00:04, 6.38MB/s]
 57%|█████▋    | 37.0M/64.9M [00:07<00:04, 6.14MB/s]
 58%|█████▊    | 37.6M/64.9M [00:07<00:04, 5.77MB/s]
 59%|█████▉    | 38.2M/64.9M [00:07<00:04, 5.69MB/s]
 60%|█████▉    | 38.9M/64.9M [00:07<00:04, 5.72MB/s]
 61%|██████    | 39.5M/64.9M [00:07<00:04, 5.85MB/s]
 62%|██████▏   | 40.2M/64.9M [00:07<00:04, 6.05MB/s]
 63%|██████▎   | 40.9M/64.9M [00:07<00:04, 6.18MB/s]
 64%|██████▍   | 41.5M/64.9M [00:07<00:04, 6.12MB/s]
 65%|██████▍   | 42.1M/64.9M [00:07<00:04, 5.76MB/s]
 66%|██████▌   | 42.8M/64.9M [00:08<00:04, 5.69MB/s]
 67%|██████▋   | 43.4M/64.9M [00:08<00:04, 5.64MB/s]
 68%|██████▊   | 44.0M/64.9M [00:08<00:03, 5.59MB/s]
 69%|██████▉   | 44.6M/64.9M [00:08<00:03, 5.55MB/s]
 70%|██████▉   | 45.2M/64.9M [00:08<00:03, 5.54MB/s]
 71%|███████   | 45.9M/64.9M [00:08<00:03, 5.50MB/s]
 72%|███████▏  | 46.5M/64.9M [00:08<00:03, 5.63MB/s]
 73%|███████▎  | 47.1M/64.9M [00:08<00:03, 5.70MB/s]
 74%|███████▎  | 47.8M/64.9M [00:09<00:03, 5.87MB/s]
 75%|███████▍  | 48.4M/64.9M [00:09<00:03, 5.65MB/s]
 75%|███████▌  | 49.0M/64.9M [00:09<00:02, 5.57MB/s]
 76%|███████▋  | 49.6M/64.9M [00:09<00:02, 5.43MB/s]
 77%|███████▋  | 50.2M/64.9M [00:09<00:02, 5.52MB/s]
 78%|███████▊  | 50.9M/64.9M [00:09<00:02, 5.62MB/s]
 79%|███████▉  | 51.5M/64.9M [00:09<00:02, 5.61MB/s]
 80%|████████  | 52.1M/64.9M [00:09<00:02, 5.65MB/s]
 81%|████████▏ | 52.8M/64.9M [00:09<00:02, 5.67MB/s]
 82%|████████▏ | 53.4M/64.9M [00:10<00:02, 5.51MB/s]
 83%|████████▎ | 54.0M/64.9M [00:10<00:02, 5.47MB/s]
 84%|████████▍ | 54.6M/64.9M [00:10<00:01, 5.55MB/s]
 85%|████████▌ | 55.2M/64.9M [00:10<00:01, 5.43MB/s]
 86%|████████▌ | 55.9M/64.9M [00:10<00:01, 5.32MB/s]
 87%|████████▋ | 56.5M/64.9M [00:10<00:01, 5.12MB/s]
 88%|████████▊ | 57.0M/64.9M [00:10<00:01, 5.10MB/s]
 89%|████████▊ | 57.5M/64.9M [00:10<00:01, 5.13MB/s]
 90%|████████▉ | 58.1M/64.9M [00:11<00:01, 5.33MB/s]
 91%|█████████ | 58.8M/64.9M [00:11<00:01, 4.19MB/s]
 92%|█████████▏| 59.9M/64.9M [00:11<00:00, 5.64MB/s]
 93%|█████████▎| 60.5M/64.9M [00:11<00:00, 5.70MB/s]
 94%|█████████▍| 61.1M/64.9M [00:11<00:00, 5.79MB/s]
 95%|█████████▌| 61.8M/64.9M [00:11<00:00, 5.81MB/s]
 96%|█████████▌| 62.4M/64.9M [00:11<00:00, 5.81MB/s]
 97%|█████████▋| 63.0M/64.9M [00:11<00:00, 5.80MB/s]
 98%|█████████▊| 63.6M/64.9M [00:12<00:00, 5.78MB/s]
 99%|█████████▉| 64.2M/64.9M [00:12<00:00, 5.68MB/s]
100%|█████████▉| 64.9M/64.9M [00:12<00:00, 5.57MB/s]
100%|██████████| 64.9M/64.9M [00:12<00:00, 5.53MB/s]

BaseOptim(
  (fixed_point): FixedPoint(
    (iterator): PGDIteration(
      (f_step): fStepPGD()
      (g_step): gStepPGD()
    )
  )
)

Evaluate the model on the problem.#

We evaluate the PnP algorithm on the test dataset, compute the PSNR metrics and plot reconstruction results.

save_folder = RESULTS_DIR / method / operation / dataset_name
plot_convergence_metrics = True  # plot metrics. Metrics are saved in save_folder.
plot_images = True  # plot images. Images are saved in save_folder.

dataloader = DataLoader(
    dataset, batch_size=batch_size, num_workers=num_workers, shuffle=False
)

test(
    model=model,
    test_dataloader=dataloader,
    physics=p,
    device=device,
    plot_images=plot_images,
    save_folder=RESULTS_DIR / method / operation / dataset_name,
    plot_convergence_metrics=plot_convergence_metrics,
    verbose=True,
)
  • Ground truth, No learning, Reconstruction
  • PSNR, F, residual
  0%|                                                                                                                           | 0/3 [00:00<?, ?it/s]
Test:   0%|                                                                                                                     | 0/3 [00:00<?, ?it/s]
Test:   0%|                                                                                   | 0/3 [00:03<?, ?it/s, PSNR=30.9, PSNR no learning=15.5]
Test:  33%|█████████████████████████                                                  | 1/3 [00:03<00:07,  3.63s/it, PSNR=30.9, PSNR no learning=15.5]
Test:  33%|█████████████████████████                                                  | 1/3 [00:03<00:07,  3.63s/it, PSNR=30.9, PSNR no learning=15.5]
Test:  33%|█████████████████████████                                                  | 1/3 [00:10<00:07,  3.63s/it, PSNR=30.3, PSNR no learning=15.8]
Test:  67%|██████████████████████████████████████████████████                         | 2/3 [00:10<00:05,  5.75s/it, PSNR=30.3, PSNR no learning=15.8]
Test:  67%|██████████████████████████████████████████████████                         | 2/3 [00:10<00:05,  5.75s/it, PSNR=30.3, PSNR no learning=15.8]
Test:  67%|██████████████████████████████████████████████████                         | 2/3 [00:13<00:05,  5.75s/it, PSNR=29.3, PSNR no learning=15.9]
Test: 100%|███████████████████████████████████████████████████████████████████████████| 3/3 [00:14<00:00,  4.73s/it, PSNR=29.3, PSNR no learning=15.9]
Test: 100%|███████████████████████████████████████████████████████████████████████████| 3/3 [00:14<00:00,  4.79s/it, PSNR=29.3, PSNR no learning=15.9]
Test results:
PSNR no learning: 15.902 +- 0.270
PSNR: 29.256 +- 1.575

{'PSNR no learning': np.float64(15.902320861816406), 'PSNR no learning_std': np.float64(0.2704836828395761), 'PSNR': np.float64(29.25592549641927), 'PSNR_std': np.float64(1.5748996872970398)}

Total running time of the script: (0 minutes 27.727 seconds)

Gallery generated by Sphinx-Gallery