Image and Visual Representation Lab

“Deepfake generation and detection is like an arms race”

Published:05.05.22 — Two EPFL computer scientists have taken home a major prize in Singapore’s Trusted Media Challenge, a five-month long competition aimed at cracking the code of deepfakes.

Our paper on self supervised learning (SSL) accepted to CVPR 2022

Published:29.03.22 — The paper "Leverage Your Local and Global Representations: A New Self-Supervised Learning Strategy" by Tong Zhang, Congpei Qiu, Wei Ke, Sabine Süsstrunk, Mathieu Salzmann is accepted to CVPR2022.

Our paper on multitask transformer will appear in CVPR 2022

Published:29.03.22 — The paper "MulT: An End-to-End Multitask Learning Transformer" by Deblina Bhattacharjee, Tong Zhang, Sabine Süsstrunk and Mathieu Salzmann is accepted to CVPR2022.

BIGPrior is published in IEEE Transactions on Image Processing

Published:10.01.22 — BIGPrior: Towards Decoupling Learned Prior Hallucination and Data Fidelity in Image Restoration by Majed El Helou and Sabine Süsstrunk published in IEEE Transactions on Image Processing.

Estimating Image Depth in the Comics Domain

Published:10.01.22 — Estimating Image Depth in the Comics Domain by Deblina Bhattacharjee, Martin Everaert, Mathieu Salzmann, Sabine Süsstrunk accepted into Winter Conference on Applications of Computer Vision (WACV 2022). 

IEEE Transactions on Neural Networks and Learning Systems (TNNLS)

Published:10.01.22 — Training Provably Robust Models by Polyhedral Envelope Regularization by Chen Liu, Mathieu Salzmann and Sabine Süsstrunk published in IEEE Transactions on Neural Networks and Learning Systems (TNNLS).

Transfer, Reconfiguration and Digital Transition in Graphic Narrative

Published:01.11.21 — Bahar Aydemir and Raphaël Baroni presented their work "Beyond the Shadow of the Z: How We Read European Bande Dessinée" in The international conference “Transfer, Reconfiguration and Digital Transition in Graphic Narratives”.

Video compression patent and licensing

Published:05.07.21 — We released an online demo of our extreme video completion method. Our technology is very energy and time-efficient on the encoder side (we require practically 0 computation and no specialized hardware like typical DCT-based encoders), it can reach extreme compression rates with complete rate flexibility and can achieve better visual reconstruction at these rates compared with the MPEG-4 H264 standard. This makes it ideal for instance for video-transmitting IoT devices, surveillance cameras, drone transmission, or emergency deployment transmission where bandwidth is very limited. Users can visualize different results and test with their own videos too on: adefan.epfl.ch For the more technically curious readers, we have an associated publication in the leading signal processing conference IEEE ICASSP that can be accessed here: https://infoscience.epfl.ch/record/277003/

Depth-guided image relighting NTIRE workshop at CVPR

Published:05.07.21 — IVRL members Majed El Helou and Ruofan Zhou co-organised this workshop, using the VIDIT dataset with geometry information. You can access the challenge paper here. 250 participants registered from various research labs, companies, and universities internationally. 

We released the image relighting dataset "VIDIT"

Published:04.07.21 — We released depth information for our image relighting dataset VIDIT. It is used for the NTIRE workshop competitions in CVPR 2021 Image relighting is fundamental for domain adaptation (data augmentation, normalization), photo montage and photo manipulation for aesthetics.