Note

New to DeepInverse? Get started with the basics with the 5 minute quickstart tutorial..

Poisson denoising using Poisson2Sparse#

This code shows how to restore a single image corrupted by Poisson noise using Poisson2Sparse, without requiring external training or knowledge of the noise level.

This method is based on the paper “Poisson2Sparse” Ta et al.[1] and restores an image by learning a sparse non-linear dictionary parametrized by a neural network using a combination of Neighbor2Neighbor Huang et al.[2], of the negative log Poisson likelihood, of the \(\ell^1\) pixel distance and of a sparsity-inducing \(\ell^1\) regularization function on the weights.

import deepinv as dinv
import torch

Load a Poisson corrupted image#

This example uses an image from the microscopy dataset FMD Zhang et al.[3].

# Seed the RNGs for reproducibility
torch.manual_seed(0)
torch.cuda.manual_seed(0)

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

physics = dinv.physics.Denoising(dinv.physics.PoissonNoise(gain=0.01, normalize=True))

x = dinv.utils.demo.load_example(
    "FMD_TwoPhoton_MICE_R_gt_12_avg50.png", img_size=(256, 256)
).to(device)
x = x[:, 0:1, :64, :64]
x = x.clamp(0, 1)
y = physics(x)

Selected GPU 0 with 3882.25 MiB free memory

Define the Poisson2Sparse model

backbone = dinv.models.ConvLista(
    in_channels=1,
    out_channels=1,
    kernel_size=3,
    num_filters=512,
    num_iter=10,
    stride=2,
    threshold=0.01,
)

model = dinv.models.Poisson2Sparse(
    backbone=backbone,
    lr=1e-4,
    num_iter=200,
    weight_n2n=2.0,
    weight_l1_regularization=1e-5,
    verbose=True,
).to(device)

Run the model#

Note that we do not pass in the physics model as Poisson2Sparse assumes a Poisson noise model internally and does not depend on the noise level.

x_hat = model(y)

# Compute and display PSNR values
learning_free_psnr = dinv.metric.PSNR()(y, x).item()
model_psnr = dinv.metric.PSNR()(x_hat, x).item()
print(f"Measurement PSNR: {learning_free_psnr:.1f} dB")
print(f"Poisson2Sparse PSNR: {model_psnr:.1f} dB")

# Plot results
dinv.utils.plot(
    [y, x_hat, x],
    titles=["Measurement", "Poisson2Sparse", "Ground truth"],
    subtitles=[f"{learning_free_psnr:.1f} dB", f"{model_psnr:.1f} dB", ""],
)
Measurement, Poisson2Sparse, Ground truth
  0%|          | 0/200 [00:00<?, ?it/s]
  0%|          | 1/200 [00:00<00:22,  8.96it/s]
  2%|▏         | 4/200 [00:00<00:10, 19.00it/s]
  4%|▎         | 7/200 [00:00<00:08, 22.62it/s]
  5%|▌         | 10/200 [00:00<00:07, 24.35it/s]
  6%|▋         | 13/200 [00:00<00:07, 25.28it/s]
  8%|▊         | 16/200 [00:00<00:07, 25.89it/s]
 10%|▉         | 19/200 [00:00<00:06, 26.29it/s]
 11%|█         | 22/200 [00:00<00:06, 26.54it/s]
 12%|█▎        | 25/200 [00:01<00:06, 26.69it/s]
 14%|█▍        | 28/200 [00:01<00:06, 26.74it/s]
 16%|█▌        | 31/200 [00:01<00:06, 26.79it/s]
 17%|█▋        | 34/200 [00:01<00:06, 26.80it/s]
 18%|█▊        | 37/200 [00:01<00:06, 26.87it/s]
 20%|██        | 40/200 [00:01<00:05, 26.90it/s]
 22%|██▏       | 43/200 [00:01<00:05, 26.90it/s]
 23%|██▎       | 46/200 [00:01<00:05, 26.93it/s]
 24%|██▍       | 49/200 [00:01<00:05, 26.96it/s]
 26%|██▌       | 52/200 [00:02<00:05, 26.99it/s]
 28%|██▊       | 55/200 [00:02<00:05, 27.02it/s]
 29%|██▉       | 58/200 [00:02<00:05, 26.69it/s]
 30%|███       | 61/200 [00:02<00:05, 26.79it/s]
 32%|███▏      | 64/200 [00:02<00:05, 26.88it/s]
 34%|███▎      | 67/200 [00:02<00:04, 26.94it/s]
 35%|███▌      | 70/200 [00:02<00:04, 26.96it/s]
 36%|███▋      | 73/200 [00:02<00:04, 27.02it/s]
 38%|███▊      | 76/200 [00:02<00:04, 26.99it/s]
 40%|███▉      | 79/200 [00:03<00:04, 27.01it/s]
 41%|████      | 82/200 [00:03<00:04, 27.04it/s]
 42%|████▎     | 85/200 [00:03<00:04, 27.09it/s]
 44%|████▍     | 88/200 [00:03<00:04, 27.11it/s]
 46%|████▌     | 91/200 [00:03<00:04, 27.14it/s]
 47%|████▋     | 94/200 [00:03<00:03, 27.15it/s]
 48%|████▊     | 97/200 [00:03<00:03, 27.15it/s]
 50%|█████     | 100/200 [00:03<00:03, 27.16it/s]
 52%|█████▏    | 103/200 [00:03<00:03, 27.14it/s]
 53%|█████▎    | 106/200 [00:04<00:03, 27.13it/s]
 55%|█████▍    | 109/200 [00:04<00:03, 27.12it/s]
 56%|█████▌    | 112/200 [00:04<00:03, 27.11it/s]
 57%|█████▊    | 115/200 [00:04<00:03, 27.08it/s]
 59%|█████▉    | 118/200 [00:04<00:03, 27.06it/s]
 60%|██████    | 121/200 [00:04<00:02, 27.02it/s]
 62%|██████▏   | 124/200 [00:04<00:02, 26.98it/s]
 64%|██████▎   | 127/200 [00:04<00:02, 26.94it/s]
 65%|██████▌   | 130/200 [00:04<00:02, 26.94it/s]
 66%|██████▋   | 133/200 [00:05<00:02, 26.93it/s]
 68%|██████▊   | 136/200 [00:05<00:02, 26.91it/s]
 70%|██████▉   | 139/200 [00:05<00:02, 26.91it/s]
 71%|███████   | 142/200 [00:05<00:02, 26.91it/s]
 72%|███████▎  | 145/200 [00:05<00:02, 26.91it/s]
 74%|███████▍  | 148/200 [00:05<00:01, 26.90it/s]
 76%|███████▌  | 151/200 [00:05<00:01, 26.90it/s]
 77%|███████▋  | 154/200 [00:05<00:01, 26.90it/s]
 78%|███████▊  | 157/200 [00:05<00:01, 26.90it/s]
 80%|████████  | 160/200 [00:06<00:01, 26.91it/s]
 82%|████████▏ | 163/200 [00:06<00:01, 26.96it/s]
 83%|████████▎ | 166/200 [00:06<00:01, 26.99it/s]
 84%|████████▍ | 169/200 [00:06<00:01, 27.01it/s]
 86%|████████▌ | 172/200 [00:06<00:01, 27.02it/s]
 88%|████████▊ | 175/200 [00:06<00:00, 27.01it/s]
 89%|████████▉ | 178/200 [00:06<00:00, 26.93it/s]
 90%|█████████ | 181/200 [00:06<00:00, 26.92it/s]
 92%|█████████▏| 184/200 [00:06<00:00, 26.85it/s]
 94%|█████████▎| 187/200 [00:07<00:00, 26.89it/s]
 95%|█████████▌| 190/200 [00:07<00:00, 26.96it/s]
 96%|█████████▋| 193/200 [00:07<00:00, 26.96it/s]
 98%|█████████▊| 196/200 [00:07<00:00, 27.02it/s]
100%|█████████▉| 199/200 [00:07<00:00, 27.04it/s]
100%|██████████| 200/200 [00:07<00:00, 26.70it/s]
Measurement PSNR: 27.5 dB
Poisson2Sparse PSNR: 30.1 dB
References:

Total running time of the script: (0 minutes 8.291 seconds)

Gallery generated by Sphinx-Gallery