Vanilla Unfolded algorithm for super-resolution

This is a simple example to show how to use vanilla unfolded Plug-and-Play. The DnCNN denoiser and the algorithm parameters (stepsize, regularization parameters) are trained jointly. For simplicity, we show how to train the algorithm on a small dataset. For optimal results, use a larger dataset.

import deepinv as dinv
import torch
from torch.utils.data import DataLoader
from deepinv.optim.data_fidelity import L2
from deepinv.optim.prior import PnP
from deepinv.unfolded import unfolded_builder
from torchvision import transforms
from deepinv.utils import get_data_home
from deepinv.datasets import BSDS500

Setup paths for data loading and results.

BASE_DIR = get_data_home()
DATA_DIR = BASE_DIR / "measurements"
RESULTS_DIR = BASE_DIR / "results"
CKPT_DIR = BASE_DIR / "ckpts"

# Set the global random seed from pytorch to ensure 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 CBSD500 dataset for training and the Set3C dataset for testing.

img_size = 64 if torch.cuda.is_available() else 32
n_channels = 3  # 3 for color images, 1 for gray-scale images
operation = "super-resolution"

Generate a dataset of low resolution images and load it.

We use the Downsampling class from the physics module to generate a dataset of low resolution images.

# For simplicity, we use a small dataset for training.
# To be replaced for optimal results. For example, you can use the larger DIV2K or LSDIR datasets (also provided in the library).

# Specify the  train and test transforms to be applied to the input images.
test_transform = transforms.Compose(
    [transforms.CenterCrop(img_size), transforms.ToTensor()]
)
train_transform = transforms.Compose(
    [transforms.RandomCrop(img_size), transforms.ToTensor()]
)
# Define the base train and test datasets of clean images.
train_base_dataset = BSDS500(
    BASE_DIR, download=True, train=True, transform=train_transform
)
test_base_dataset = BSDS500(
    BASE_DIR, download=False, train=False, transform=test_transform
)

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

# Degradation parameters
factor = 2
noise_level_img = 0.03

# Generate the gaussian blur downsampling operator.
physics = dinv.physics.Downsampling(
    filter="gaussian",
    img_size=(n_channels, img_size, img_size),
    factor=factor,
    device=device,
    noise_model=dinv.physics.GaussianNoise(sigma=noise_level_img),
)
my_dataset_name = "demo_unfolded_sr"
n_images_max = (
    None if torch.cuda.is_available() else 10
)  # max number of images used for training (use all if you have a GPU)
measurement_dir = DATA_DIR / "BSDS500" / operation
generated_datasets_path = dinv.datasets.generate_dataset(
    train_dataset=train_base_dataset,
    test_dataset=test_base_dataset,
    physics=physics,
    device=device,
    save_dir=measurement_dir,
    train_datapoints=n_images_max,
    num_workers=num_workers,
    dataset_filename=str(my_dataset_name),
)

train_dataset = dinv.datasets.HDF5Dataset(path=generated_datasets_path, train=True)
test_dataset = dinv.datasets.HDF5Dataset(path=generated_datasets_path, train=False)

0it [00:00, ?it/s]
832kB [00:00, 7.87MB/s]
5.56MB [00:00, 31.5MB/s]
8.69MB [00:00, 29.8MB/s]
11.6MB [00:00, 29.2MB/s]
14.5MB [00:00, 29.1MB/s]
17.3MB [00:00, 28.6MB/s]
20.1MB [00:00, 28.3MB/s]
23.2MB [00:00, 29.7MB/s]
26.5MB [00:00, 30.9MB/s]
29.5MB [00:01, 30.8MB/s]
32.9MB [00:01, 32.1MB/s]
36.0MB [00:01, 31.9MB/s]
39.3MB [00:01, 32.6MB/s]
42.6MB [00:01, 32.9MB/s]
45.9MB [00:01, 33.2MB/s]
49.1MB [00:01, 31.7MB/s]
52.4MB [00:01, 32.7MB/s]
55.6MB [00:01, 31.6MB/s]
58.7MB [00:02, 24.6MB/s]
61.3MB [00:02, 22.3MB/s]
63.6MB [00:02, 20.8MB/s]
65.8MB [00:02, 19.6MB/s]
67.8MB [00:02, 19.0MB/s]
69.6MB [00:02, 18.7MB/s]
71.5MB [00:02, 18.3MB/s]
73.3MB [00:02, 17.7MB/s]
75.1MB [00:03, 17.7MB/s]
76.8MB [00:03, 17.7MB/s]
78.6MB [00:03, 16.9MB/s]
80.5MB [00:03, 17.0MB/s]
82.6MB [00:03, 18.2MB/s]
84.4MB [00:03, 18.1MB/s]
86.2MB [00:03, 17.8MB/s]
88.0MB [00:03, 18.0MB/s]
89.8MB [00:03, 17.8MB/s]
91.5MB [00:04, 18.0MB/s]
93.2MB [00:04, 17.9MB/s]
95.0MB [00:04, 17.7MB/s]
96.8MB [00:04, 17.9MB/s]
98.5MB [00:04, 17.9MB/s]
100MB [00:04, 17.8MB/s]
102MB [00:04, 17.8MB/s]
104MB [00:04, 17.5MB/s]
106MB [00:04, 17.4MB/s]
107MB [00:05, 17.4MB/s]
109MB [00:05, 17.7MB/s]
111MB [00:05, 17.9MB/s]
113MB [00:05, 17.9MB/s]
114MB [00:05, 17.6MB/s]
116MB [00:05, 17.5MB/s]
118MB [00:05, 17.3MB/s]
119MB [00:05, 17.3MB/s]
121MB [00:05, 17.4MB/s]
123MB [00:05, 16.4MB/s]
125MB [00:06, 16.7MB/s]
126MB [00:06, 16.7MB/s]
128MB [00:06, 17.3MB/s]
130MB [00:06, 17.0MB/s]
132MB [00:06, 17.9MB/s]
134MB [00:06, 9.28MB/s]
141MB [00:07, 19.8MB/s]
144MB [00:07, 18.4MB/s]
149MB [00:07, 24.9MB/s]
153MB [00:07, 27.5MB/s]
156MB [00:07, 15.8MB/s]
159MB [00:08, 12.3MB/s]
160MB [00:08, 19.9MB/s]

Extracting:   0%|          | 0/2492 [00:00<?, ?it/s]
Extracting:  13%|█▎        | 329/2492 [00:00<00:00, 3288.83it/s]
Extracting:  26%|██▋       | 658/2492 [00:00<00:00, 3288.10it/s]
Extracting:  40%|███▉      | 989/2492 [00:00<00:00, 3294.80it/s]
Extracting:  53%|█████▎    | 1319/2492 [00:00<00:00, 1578.77it/s]
Extracting:  62%|██████▏   | 1554/2492 [00:00<00:00, 1290.30it/s]
Extracting:  70%|██████▉   | 1736/2492 [00:01<00:00, 1242.80it/s]
Extracting: 100%|██████████| 2492/2492 [00:01<00:00, 2114.40it/s]
Dataset has been saved at datasets/measurements/BSDS500/super-resolution/demo_unfolded_sr0.h5

Define the unfolded PnP algorithm.

We use the helper function deepinv.unfolded.unfolded_builder() to define the Unfolded architecture. The chosen algorithm is here DRS (Douglas-Rachford Splitting). Note that if the prior (resp. a parameter) is initialized with a list of length max_iter, then a distinct model (resp. parameter) is trained for each iteration. For fixed trained model prior (resp. parameter) across iterations, initialize with a single element.

# Unrolled optimization algorithm parameters
max_iter = 5  # number of unfolded layers

# Select the data fidelity term
data_fidelity = L2()

# Set up the trainable denoising prior
# Here the prior model is common for all iterations
prior = PnP(denoiser=dinv.models.DnCNN(depth=7, pretrained=None).to(device))

# The parameters are initialized with a list of length max_iter, so that a distinct parameter is trained for each iteration.
stepsize = [1.0] * max_iter  # stepsize of the algorithm
sigma_denoiser = [
    1.0
] * max_iter  # noise level parameter of the denoiser (not used by DnCNN)
beta = 1.0  # relaxation parameter of the Douglas-Rachford splitting
params_algo = {  # wrap all the restoration parameters in a 'params_algo' dictionary
    "stepsize": stepsize,
    "g_param": sigma_denoiser,
    "beta": beta,
}
trainable_params = [
    "stepsize",
    "beta",
]  # define which parameters from 'params_algo' are trainable

# Logging parameters
verbose = True

# Define the unfolded trainable model.
model = unfolded_builder(
    iteration="DRS",
    params_algo=params_algo.copy(),
    trainable_params=trainable_params,
    data_fidelity=data_fidelity,
    max_iter=max_iter,
    prior=prior,
)

Define the training parameters.

We use the Adam optimizer and the StepLR scheduler.

# training parameters
epochs = 5 if torch.cuda.is_available() else 2
learning_rate = 5e-4
train_batch_size = 32 if torch.cuda.is_available() else 1
test_batch_size = 3

# choose optimizer and scheduler
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate, weight_decay=1e-8)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=int(epochs * 0.8))

# choose supervised training loss
losses = [dinv.loss.SupLoss(metric=dinv.metric.MSE())]

train_dataloader = DataLoader(
    train_dataset, batch_size=train_batch_size, num_workers=num_workers, shuffle=True
)
test_dataloader = DataLoader(
    test_dataset, batch_size=test_batch_size, num_workers=num_workers, shuffle=False
)

Train the network

We train the network using the deepinv.Trainer class.

trainer = dinv.Trainer(
    model,
    physics=physics,
    train_dataloader=train_dataloader,
    eval_dataloader=test_dataloader,
    epochs=epochs,
    scheduler=scheduler,
    losses=losses,
    optimizer=optimizer,
    device=device,
    early_stop=True,  # set to None to disable early stopping
    save_path=str(CKPT_DIR / operation),
    verbose=verbose,
    show_progress_bar=False,  # disable progress bar for better vis in sphinx gallery.
)

model = trainer.train()

The model has 188169 trainable parameters
Train epoch 0: TotalLoss=0.337, PSNR=6.206
Eval epoch 0: PSNR=9.392
Best model saved at epoch 1
Train epoch 1: TotalLoss=0.125, PSNR=10.181
Eval epoch 1: PSNR=10.399
Best model saved at epoch 2

Test the network

trainer.test(test_dataloader)

test_sample, _ = next(iter(test_dataloader))
model.eval()
test_sample = test_sample.to(device)

# Get the measurements and the ground truth
y = physics(test_sample)
with torch.no_grad():
    rec = model(y, physics=physics)

backprojected = physics.A_adjoint(y)

dinv.utils.plot(
    [backprojected, rec, test_sample],
    titles=["Linear", "Reconstruction", "Ground truth"],
    suptitle="Reconstruction results",
)

Eval epoch 0: PSNR=10.399, PSNR no learning=9.623
Test results:
PSNR no learning: 9.623 +- 3.848
PSNR: 10.399 +- 3.451
/home/runner/work/deepinv/deepinv/deepinv/utils/plotting.py:379: UserWarning: This figure was using a layout engine that is incompatible with subplots_adjust and/or tight_layout; not calling subplots_adjust.
  fig.subplots_adjust(top=0.75)