Lightnews — Scholar-powered news

Julian Tachella

@tachellajulian.bsky.social

260 followers 140 following 68 posts

CNRS research scientist, based at ENS de Lyon I'm interested in AI for imaging inverse problems Looking to hire phds/postdocs! 🇦🇷🇬🇧🇫🇷 Website: https://tachella.github.io/ Deepinverter: https://deepinv.github.io/

tachella.github.io

Posts Media Videos Starter Packs

Pinned

Julian Tachella @tachellajulian.bsky.social · Mar 17

🚀 Do you want to learn about self-supervised learning for inverse problems?

▶️ Check out the 3-hour tutorial presented by Mike Davies (University of Edinburgh) and myself (CNRS, ENS Lyon), covering all the recent advances in a unified and simplified framework!

youtube.com/playlist?lis...

Self-Supervised Learning for Inverse Problems - YouTube

Tutorial by Julián Tachella (CNRS, ENS Lyon) & Mike Davies (University of Edinburgh) given at the University of Edinburgh, February 2025.I. "Introduction"II....

youtube.com

1 4

Julian Tachella @tachellajulian.bsky.social · 19m

Read more about this new feature in the docs: deepinv.github.io/deepinv/user...

and check out a Jupyter notebook demo: deepinv.github.io/deepinv/auto...

This is a great example of the under-the-hood features that make large-scale training seamless with the library!

Unfolded Algorithms — deepinv 0.3.4 documentation

deepinv.github.io

Julian Tachella @tachellajulian.bsky.social · 19m

The new feature leverages closed-form formulas for computing gradients without storing intermediate steps, only requiring an additional call to the least squares solver.

Julian Tachella @tachellajulian.bsky.social · 19m

DeepInverse provided access to state-of-the-art matrix-free least squares solvers (eg, conjugate gradient, BiCGStab, lsqr, etc), but backproping through their steps required significant memory and slowed down training.

Julian Tachella @tachellajulian.bsky.social · 19m

Many reconstruction models rely on a differentiable least squares solver, such as unfolded networks with proximal steps or reconstruct-anything-model.

Julian Tachella @tachellajulian.bsky.social · 19m

🔎 A focus on the new implicit backprop for least squares solvers (by Minh Hai Nguyen), which unlocks training in large-scale imaging settings:

Unfolded Algorithms — deepinv 0.3.4 documentation

deepinv.github.io

Julian Tachella @tachellajulian.bsky.social · 19m

🚀 New Features:

- Implicit backprop for least squares solvers

- noise statistics for SAR imaging

- Multi-coil MRI coil-map estimation acceleration via CuPy

- RicianNoise model

- Self-supervised super-resolution loss

Unfolded Algorithms — deepinv 0.3.4 documentation

deepinv.github.io

Julian Tachella @tachellajulian.bsky.social · 19m

💥 DeepInverse v0.3.5 is out! 💥

docs: deepinv.github.io/deepinv/

github: github.com/deepinv/deep...

DeepInverse: a Python library for imaging with deep learning — deepinv 0.3.4 documentation

deepinv.github.io

1 1 3

Julian Tachella @tachellajulian.bsky.social · 28d

Stay tuned for more hackathons to come!
🫂If you are interested in joining the community
- join the discord discord.gg/qBqY5jKw3p
- open an issue or a pull request deepinv.github.io
- drop us a message!

Join the deepinv library Discord Server!

Check out the deepinv library community on Discord – hang out with 189 other members and enjoy free voice and text chat.

discord.gg

Julian Tachella @tachellajulian.bsky.social · 28d

❔ Uncertainty Quantification
- coverage plots + conformal prediction
🏫 Trainer v2!

Julian Tachella @tachellajulian.bsky.social · 28d

💥 Self-supervised learning
- Noise2Self
- MERLIN for SAR

Julian Tachella @tachellajulian.bsky.social · 28d

📖 Datasets
- PETRIC challenge dataset
- BSD500 dataset
- Calgary MRI dataset

Julian Tachella @tachellajulian.bsky.social · 28d

🚀 More optimizers
- implicit differentiation for linear solvers
- more noisy data fidelities for diffusion
- SPECT preconditioned reconstruction
- phase unwrapping reconstruction

Julian Tachella @tachellajulian.bsky.social · 28d

🖼️ More image priors/regularizers
- latent diffusion models
- general restoration models for PnP
- complex wavelets prior
- weakly convex regularisers

Julian Tachella @tachellajulian.bsky.social · 28d

🌌 Large-scale imaging:
- multi-GPU distributed denoising/reconstruction
- 3D TV/TGV denoising in example
- 3D DRUNet

Julian Tachella @tachellajulian.bsky.social · 28d

🔬 New forward operators
- ultrasound physics
- MRI-NUFFT physics wrapper
- ++ ASTRA integration
- near & far field radar operators
- pytomography wrapper SPECT
- SAR noise models
- multiview operators
- single-pixel spyrit wrapper
- spatial unwrapping operators
- multiscale

1 1

Julian Tachella @tachellajulian.bsky.social · 28d

☀️ Just wrapped up the DeepInverse Hackathon!

We had 30+ imaging scientists from all over the world coding during 3 days next to the beautiful Calanques in Marseille, France. It was a great moment to meet new people, discuss science, and code new imaging algorithms!

1 3 11

Julian Tachella @tachellajulian.bsky.social · Aug 20

Also available on Google Colab colab.research.google.com/drive/11YKc_...

Julian Tachella @tachellajulian.bsky.social · Aug 20

For every step, you can either i) use a preexisting {operator,model,dataset} or ii) define a custom one yourself.

Julian Tachella @tachellajulian.bsky.social · Aug 20

The tutorial, built by @andrewwango.bsky.social, walks you through the main steps of a computational imaging pipeline:

📸 Defining your forward operator
💻 Defining your reconstruction model
🧾 Defining your dataset

Julian Tachella @tachellajulian.bsky.social · Aug 20

🔬 Are you working in computational imaging or interested in learning?

📖 DeepInverse just released a new 5-minute quickstart tutorial

deepinv.github.io/deepinv/auto...

which gets you started developing AI models for reconstructing images!

1 2 3

Julian Tachella @tachellajulian.bsky.social · May 19

🔗 Worked-out example: deepinv.github.io/deepinv/auto...

🔗 Diffusion user guide: deepinv.github.io/deepinv/user...

Building your diffusion posterior sampling method using SDEs — deepinverse 0.3 documentation

deepinv.github.io

Julian Tachella @tachellajulian.bsky.social · May 19

🙏 This is thanks to the contribution of Minh Hai Nguyen

Julian Tachella @tachellajulian.bsky.social · May 19

We also maintain black-box implementations of popular methods like DPS, DDRM, and DiffPIR—ready to use out of the box.

Julian Tachella @tachellajulian.bsky.social · May 19

New Feature in DeepInverse (deepinv.github.io):

🚀 Custom Diffusion Solver Design
DeepInverse now simplifies the process with:

✔ Standard SDEs (VP, VE, etc.)
✔ Pretrained denoisers for multiple noise levels
✔ ODE/SDE solvers (Euler, Heun)
✔ Noisy data fidelity terms for guidance

Redirecting to https://deepinv.github.io/deepinv/

deepinv.github.io

2 6

Julian Tachella @tachellajulian.bsky.social · Apr 21

I will also give a 1-hour talk at NUS (School of Computing) on Wednesday 10 am, see more info here: events.comp.nus.edu.sg/view/24032,

🖇️ code : github.com/tachella/uns...
🖇️ paper: openreview.net/forum?id=ScV...