Julián Tachella

julian.tachella@cnrs.fr

I am a research scientist at the French National Centre for Scientific Research CNRS, working at the Sisyph laboratory, École Normale Supérieure de Lyon (Lyon, France). My research lies at the intersection of signal processing and machine learning. I am particularly interested in the theory of imaging inverse problems and applications in computational imaging.

Before joining the CNRS, I did a postdoc in Mike Davies group at the University of Edinburgh (Edinburgh, UK) and a PhD at Heriot-Watt University (Edinburgh, UK) and TéSA lab (Toulouse, France) under the supervision of Yoann Altmann, Jean-Yves Tourneret and Steve McLaughlin.

Code and videos related to my publications can be found here.

news

Sep 17, 2022	Our paper on unsupervised learning from incomplete measurements has been accepted at NeurIPS 2022.
Sep 9, 2022	Our review paper on equivariant deep learning for imaging has been accepted at IEEE Signal Processing Magazine.
May 27, 2022	Our paper received the Best Student Paper Award at ICASSP’22!
Mar 2, 2022	Our paper on robust equivariant imaging is accepted at CVPR’22 (Oral!)
Jan 1, 2022	Accepted paper on 3D reconstruction with sketching at ICASSP’22

selected publications

Equivariant Imaging: Learning Beyond the Range Space

Chen, Dongdong, Tachella, Julian, and Davies, Mike E

In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Mar 2021

Abs PDF Video Project Code

In various imaging problems, we only have access to compressed measurements of the underlying signals, hindering most learning-based strategies which usually require pairs of signals and associated measurements for training. Learning only from compressed measurements is impossible in general, as the compressed observations do not contain information outside the range of the forward sensing operator. We propose a new end-to-end self-supervised framework that overcomes this limitation by exploiting the equivariances present in natural signals. Our proposed learning strategy performs as well as fully supervised methods. Experiments demonstrate the potential of this framework on inverse problems including sparse-view X-ray computed tomography on real clinical data and image inpainting on natural images.
The Neural Tangent Link Between CNN Denoisers and Non-Local Filters

Tachella, Julian, Tang, Junqi, and Davies, Mike

In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Jun 2021

Abs PDF Blog Video Project Code

Convolutional Neural Networks (CNNs) are now a well-established tool for solving computational imaging problems. Modern CNN-based algorithms obtain state-of-the-art performance in diverse image restoration problems. Furthermore, it has been recently shown that, despite being highly overparameterized, networks trained with a single corrupted image can still perform as well as fully trained networks. We introduce a formal link between such networks through their neural tangent kernel (NTK), and well-known non-local filtering techniques, such as non-local means or BM3D. The filtering function associated with a given network architecture can be obtained in closed form without need to train the network, being fully characterized by the random initialization of the network weights. While the NTK theory accurately predicts the filter associated with networks trained using standard gradient descent, our analysis shows that it falls short to explain the behaviour of networks trained using the popular Adam optimizer. The latter achieves a larger change of weights in hidden layers, adapting the non-local filtering function during training. We evaluate our findings via extensive image denoising experiments.
Real-time 3D reconstruction from single-photon lidar data using plug-and-play point cloud denoisers

Tachella, Julian, Altmann, Yoann, Mellado, Nicolas, Tobin, Rachel, McCarthy, Angus, Buller, Gerald, Tourneret, Jean-Yves, and McLaughlin, Steve

Nature Communications Jun 2019

Abs PDF Video Project Code

Single-photon lidar has emerged as a prime candidate technology for depth imaging through challenging environments. Until now, a major limitation has been the significant amount of time required for the analysis of the recorded data. Here we show a new computational framework for real-time three-dimensional (3D) scene reconstruction from single-photon data. By combining statistical models with highly scalable computational tools from the computer graphics community, we demonstrate 3D reconstruction of complex outdoor scenes with processing times of the order of 20 ms, where the lidar data was acquired in broad daylight from distances up to 320 metres. The proposed method can handle an unknown number of surfaces in each pixel, allowing for target detection and imaging through cluttered scenes. This enables robust, real-time target reconstruction of complex moving scenes, paving the way for single-photon lidar at video rates for practical 3D imaging applications.