Stage Recherche Innovation IA (TensorFlow,PyTorch,algorithms)

If you have ever watched a television show or live event or enjoyed a movie in VOD on your phone or tablet this experience was likely brought to you through an Ateme solution.

We are Ateme (PARIS: ATEME). We are the video delivery leader helping leading content providers service providers and pure streaming players boost their engagement acquire new viewers and create new sources of revenues. Leveraging our continuous investment in R&D and innovation we empower our customers to deliver a high quality of experience on any screen.

Delivering video experiences also has an impact on our world. Thats why our multiple award-winning engineering teams design efficient and flexible solutions that cut waste with no compromise on quality. So that viewers can enjoy their unique experiences and the world we live in well into the future.

Thanks to a strong CSR policy that reinforces our mission to Make the entertainment and video experience captivating greener and accessible to everyone we strive every day to build a better and more sustainable future for our planet our people and our ecosystem.

At Ateme we value innovation collaboration empowerment agility and everyones contributions. We offer cross-culture enrichment thanks to employees of 30 different nationalities. We consider the globe as our playground and we facilitate mobility internationally especially in our offices in France Sao Paulo Denver New York and Singapore.

Be part of our team and join our fantastic journey!

Main activities:

Subject: Exploring quality metrics for Neural Video Codec

Nowadays our societies are highly dependent on digital video broadcast. An industry which is itself highly restricted by the bandwidth for distribution of a tremendous amount of video over different transportation means (Terrestrial Satellite Internet mobile). The increase in video consumption leads to a massive usage of computational resources in datacenters. To deal with this increase several video compression algorithms have been standardized such as Advanced Video Coding (AVC)1 High Efficiency Video Coding (HEVC) 2 and the latest video codec Versatile Video Coding (VVC)3.

However In the past few years these traditional codecs have been challenged by neural video fact end-to-end neural image and video coding approaches 4 5 6 have shown promising results making them compete with HEVC and VVC in terms of compression efficiency. These video coding models minimize a loss function that includes a distortion term between source and decoding frames. Currently classic metrics are used such as MSE/ PSNR and MS-SSIM. The goal of this internship is to explore additional quality metrics. Several categories exist in the literature such as Full-reference perceptual and blind metrics. The idea is to integrate them into both the training and the evaluation process of the video coding model particularly within the loss function and the quality assessment computations.

This internship will consist of following phases:

1. State-of-the art of video quality metrics.

1. Introduction understanding and operating with the existing neural video codec.

1. Integration of selected quality metrics in the training and evaluation process of the neural codec.

1. Analysis and comparison of different models across multiple metrics.

In all above phases the candidate will constantly interact with his/her colleagues to benefit from their knowledge and experience. Noteworthy of precising that based on the profile of the candidate the internship orientation might be adjusted.

References :

1 T. Wiegand GJ Sullivan G. Bjontegaard and A Luthra. 2003. Overview of the H. 264/AVC video coding standard. IEEE Transactions on circuits and systems for video technology 13 7 (2003) 560576.

2 G. Sullivan J. Ohm W. Han and T. Wiegand. 2012. Overview of the high efficiency video coding (HEVC) standard. IEEE Transactions on circuits and systems for video technology 22 12 (2012)

3 G. Sullivan et J. R. Ohm Versatile video coding Towards the next generation of video compression chez Picture Coding Symposium 2018

4 B. L. Y. L. Jiahao Li Hybrid Spatial-Temporal Entropy Modelling for Neural Video Compression Proceedings of the 30th ACM International Conference on Multimedia p. 15.

5 J. Li B. Li et Y. Lu Neural video compression with diverse contexts. chez CVPR 2023.

6 H. M. Kwan G. Gao F. Zhang A. Gower Bull et D HiNeRV: Video Compression with Hierarchical Encoding-based Neural Representation. chez Advances in Neural Information Processing Systems 2024.

About the candidate:

• Last-year student of engineering school program or masters in computer science or electrical engineering.

• Excellent python programming skills.

• Hand-on experience with neural networks and libraries such as TensorFlow PyTorch etc.

• Ability to read and understand academic papers in the domain of image and video processing.

• Being familiar with video compression algorithms is a plus.

• Fluent in English.

Location: The position is based in Rennes (35)

Benefits: 1500 gross Ticket restaurant Reimbursement of transportation fees possibility of recruitment (CDI) or CIFRE thesis

Contacts:

To find out more about our company do not hesitate to visit our website: