#### NOTE New to DeepInverse? Get started with the basics with the [5 minute quickstart tutorial](https://deepinv.org/auto_examples/basics/demo_quickstart.html.md#sphx-glr-auto-examples-basics-demo-quickstart-py).. # Use a pretrained model Follow this example to reconstruct images using a pretrained model in one line. We show three sets of general pretrained reconstruction methods, including: * Pretrained feedforward [`Reconstruct Anything Model (RAM)`](https://deepinv.org/api/stubs/deepinv.models.RAM.html.md#deepinv.models.RAM); * [Plug-and-play](https://deepinv.org/user_guide/reconstruction/iterative.html.md#iterative) with a pretrained denoiser. * Pretrained [diffusion model](https://deepinv.org/user_guide/reconstruction/sampling.html.md#diffusion); See [pretrained models](https://deepinv.org/user_guide/reconstruction/pretrained-models.html.md#pretrained-models) for a principled comparison between methods demonstrated in this example. #### TIP * Want to use your own dataset? See [Bring your own dataset](https://deepinv.org/auto_examples/basics/demo_custom_dataset.html.md#sphx-glr-auto-examples-basics-demo-custom-dataset-py) * Want to use your own physics? See [Bring your own physics](https://deepinv.org/auto_examples/basics/demo_custom_physics.html.md#sphx-glr-auto-examples-basics-demo-custom-physics-py) ```Python import deepinv as dinv import torch device = dinv.utils.get_device() ``` ```none Selected GPU 0 with 28686.0 MiB free memory ``` Let’s say you want to reconstruct a butterfly from noisy, blurry measurements: ```Python # Ground truth x = dinv.utils.load_example("butterfly.png", device=device) # Define physics physics = dinv.physics.BlurFFT( x.shape[1:], filter=dinv.physics.functional.gaussian_blur(sigma=(5, 5)), noise_model=dinv.physics.GaussianNoise( sigma=0.1, rng=torch.Generator(device=x.device).manual_seed(123) ), device=device, ) y = physics(x) ``` For each model, define model in one line and reconstruct in one line. Pretrained Reconstruct Anything Model: #### SEE ALSO See [Inference and fine-tune a foundation model](https://deepinv.org/auto_examples/models/demo_foundation_model.html.md#sphx-glr-auto-examples-models-demo-foundation-model-py) for further one-line examples for the RAM model across various domains. ```Python model = dinv.models.RAM(pretrained=True, device=device) with torch.no_grad(): x_hat1 = model(y, physics) ``` PnP algorithm with pretrained denoiser: #### SEE ALSO See [pretrained denoisers](https://deepinv.org/user_guide/reconstruction/pretrained-models.html.md#pretrained-weights) for a full list of denoisers that can be plugged into iterative/sampling algorithms. See [DPIR method for PnP image deblurring.](https://deepinv.org/auto_examples/plug-and-play/demo_PnP_DPIR_deblur.html.md#sphx-glr-auto-examples-plug-and-play-demo-pnp-dpir-deblur-py) for a further example of using plug-and-play with a pretrained denoiser. ```Python denoiser = dinv.models.DRUNet(device=device) model = dinv.optim.DPIR(sigma=0.1, denoiser=denoiser, device=device) x_hat2 = model(y, physics) ``` Pretrained diffusion model (we reduce the image size for demo speed on CPU, as diffusion model is slow): #### SEE ALSO See [Image reconstruction with a diffusion model](https://deepinv.org/auto_examples/sampling/demo_ddrm.html.md#sphx-glr-auto-examples-sampling-demo-ddrm-py) for a further example of using a pretrained diffusion model. ```Python model = dinv.sampling.DDRM(denoiser, sigmas=torch.linspace(1, 0, 20)).to(device) x_hat3 = model(y, physics) ``` Plot results ```Python dinv.utils.plot( { "Ground Truth": x, "Measurement": y, "Pretrained \nRAM": x_hat1, "Pretrained \nPnP": x_hat2, "Pretrained \nDiffusion": x_hat3, }, subtitles=[ "PSNR:", f"{dinv.metric.PSNR()(y, x).item():.2f} dB", f"{dinv.metric.PSNR()(x_hat1, x).item():.2f} dB", f"{dinv.metric.PSNR()(x_hat2, x).item():.2f} dB", f"{dinv.metric.PSNR()(x_hat3, x).item():.2f} dB", ], figsize=(10, 5), ) ``` 🎉 Well done, you now know how to use pretrained models! ## What’s next? * Check out the [example on how to fine-tune a foundation model to your own problem](https://deepinv.org/auto_examples/models/demo_foundation_model.html.md#sphx-glr-auto-examples-models-demo-foundation-model-py). * See [pretrained models](https://deepinv.org/user_guide/reconstruction/pretrained-models.html.md#pretrained-models) for a comparison between methods demonstrated in this example. * See [diffusion](https://deepinv.org/user_guide/reconstruction/sampling.html.md#diffusion) and [iterative](https://deepinv.org/user_guide/reconstruction/iterative.html.md#iterative) for how to fully customize your sampling or iterative algorithm using a pretrained denoiser. **Total running time of the script:** (0 minutes 1.123 seconds) 🏷 Tags: Denoising [`Download Jupyter notebook: demo_pretrained_model.ipynb`](https://deepinv.org/auto_examples/basics/_downloads/7bd3f0c37fab9846872188d9d92e6b7e/demo_pretrained_model.ipynb) [`Download Python source code: demo_pretrained_model.py`](https://deepinv.org/auto_examples/basics/_downloads/d2c8ff79555a2ba421858b0a406e211e/demo_pretrained_model.py) [`Download zipped: demo_pretrained_model.zip`](https://deepinv.org/auto_examples/basics/_downloads/df43f8c0a78b3a0cbb70c84c0476beb6/demo_pretrained_model.zip) [Gallery generated by Sphinx-Gallery](https://sphinx-gallery.github.io)