[CLOSED] [IML_CV_DM_021] Deep 3D Object Localization & Tracking for Real / Virtual Fusion 

To fuse virtual objects with real content (or the opposite) it is necessary to know the parameters of the virtual and real cameras. In particular, the 3D localization of the real camera and the real object need to be computed and tracked to ensure temporal consistency and fast computing. Current solutions rely on visual pattern tracking. More recent technologies consider fusing information from various sensors as used for automotive applications, such as accelerometer, gyroscope, GPS, LIDAR, radar… The goal of the internship is to improve current algorithms and propose innovative solutions for filtering and fusing the various signals. Deep learning could also be considered as an improvement of existing techniques

Skills : Applied Mathematics, Signal Processing, Programming, Computer Vision, Machine/Deep Learning

Keywords : localization, sensors, filtering, machine learning, VFX, movies

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[CLOSED] [ISL_CV_DM_VP_022] Framework for Deep Learning Inference 

Deep learning approach enable very efficient signal processing and pattern recognition tasks in several fields including image segmentation or 3D modelling. Many frameworks are available for designing and training deep model. But the deployment of deep based inference solution in professional environments is still difficult because of many integration constraints (eg GPU, interactivity, resource sharing and complexity of DL libraries). The goal of this internship is to develop a library and a workflow for transferring, converting and using deep models into final VFX production environments. A possible use case is the interactive object segmentation

Skills : machine learning, deep learning, computer vision, video/image processing, PyTorch, TensorFlow or Keras deep learning frameworks, Python or C++. Cuda, OpenCL

Keywords : machine learning (deep learning), video processing, computer vision, segmentation

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[CLOSED] [ISL_CV_DM_VP_023] Deep learning for rotoscoping task 

Very recently deep learning approaches allowed developing very efficient approaches in various fields (e.g., image/video processing, computer vision, audio processing). This internship proposal targets development of deep learning approaches for high-end visual effects. The approaches may address problems such as image segmentation, rotoscoping, matting, despilling and tracking; and might be integrated in a professional VFX software to help the colorists

Skills : machine learning, deep learning, computer vision, video/image processing, PyTorch, TensorFlow or Keras deep learning frameworks, Python or C++.

Keywords : machine learning (deep learning), video processing, computer vision, segmentation, tracking, rotoscoping, matting

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[CLOSED] [ISL_DM_VP_024] Deep learning-based filtering for future video codec 

The topic of this internship is the image (or video) compression using deep learning approaches. The candidate will have the opportunity to work on state-of –the art video coding technology and improve some crucial parts of a video codec using CNN based filtering. Optimization of complexity and rate-distortion criteria are also part of the challenge. The candidate should be familiar with current deep learning software packages and have a good background in image processing in general

Skills : deep learning algorithms and software, programming (C++/python), image processing (denoising, deblocking etc.), video compression

Keywords : image restoration, deep learning, video compression

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[CLOSED] [ISL_CG_CV_DM_027] Weather Reassignment 

In many places throughout the world weathercams are installed. Each camera creates timelapse photography of an environment over different times of day, atmospheric conditions, and seasons. This data could be leveraged to train models, enabling detection of different types of weather and time of day. Secondly, neural networks could be trained using this data to then apply new atmospheric conditions, weather conditions, seasonal change etc., to new images. The purpose of this research would be to understand if such image transformations are possible and to develop tools to perform such transformations. Eventually, the aim is to support our special effects businesses with new tools to carry out high level image and video edits

Skills : machine learning, 3d animation, Python, C++.

Keywords : appearance transfer, weathercam, deep learning

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[CLOSED] [ISL_CV_DM_028] Weakly supervised quadruped modelling from video  

The goal of this internship is to develop a novel weakly supervised deep learning approach to extraction quadrupeds’ motion parameters from unconstrained videos. The motivation for this work is to build novel probabilistic models of the motion for several classes of quadrupeds. Applications include synthetic animation as well as video analysis and retargeting

Skills : machine learning, Python/C++, GPU

Keywords : deep learning, optimization, CV/CG

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[CLOSED] [ISL_CV_DM_029] Self-supervised deep solvers for optimization-based image manipulation 

In different domains, deep convolutional networks trained with self-supervisions have appeared as a powerful alternative to classic solvers. They allow fast runtime, circumvent initialization issues and, might yield better results in certain regimes. This internship aims at developing further this promising type of approach in the specific context of photo-realistic example-based image manipulation

Skills : machine learning, Python/C++, GPU

Keywords : deep learning, optimization, image editing

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[OPEN] [ISL_CV_DM_030] Wasserstein embedding learning for animation data 

Defining relevant metric between animation data is crucial for many applications. For that purpose, Wasserstein distances are a better, but more complex, alternative to Euclidian ones. We propose in this internship to learn deep embeddings where the Euclidian distance in the output space mimics the Wasserstein one in the original space of animation data

Skills : machine learning, Python/C++, GPU

Keywords : deep learning, optimization, CV/CG

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[CLOSED] [IML_CV_DM_031] Deep Learning for Light-Fields 

Deep learning has already shown a big potential in many computer vision tasks. In particular it has been shown recently that depth estimation and new view synthesis for Light-Fields are two highly related problems that can be solved with deep learning techniques. However, those algorithms have only been tested on a type of Light-Field capturing devices (plenoptic cameras) having a very small parallax. In this context, this internship aims at further developing these promising approaches to other Light-Field data such as multi-camera datasets. The internship will consist in studying the state-of-the-art in deep-learning for depth-estimation and view synthesis, the implementation of a deep-learning solution specially tailored for Light-Fields and the training, validation and testing on different datasets

Skills : Machine learning, Computer vision, Python

Keywords : Deep learning, Light-Fields, Depth estimation, View synthesis

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[CLOSED] [ISL_DM_036] Audio attributes modification with deep representations 

Very recently deep learning approaches allowed developing very efficient approaches for modification of image attributes (for example for faces: age, wearing eyeglasses, moustache). This internship proposal targets the development of similar approaches for modification of audio attributes (for example for speech: speaker’s age, identity, accent). These new approaches may be applied for speech and/or general audio manipulation tasks

Skills : machine learning, audio processing, speech processing, Python, Matlab or C++

Keywords : machine learning (deep learning), audio processing, speech processing

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[CLOSED] [IML_CG_DM_037] Style transfer for VR avatar 

Thanks to the affordable 3D scanners, users can now create a photorealistc 3D model of their body which can be used in VR environment. But this model may not be suitable for any content. The style, color, shape, etc. may not be adapted. In this internship, we would like to design a system to automatically adapt a given 3D model to a target VR content.

Skills : 3D Engine, Computer Graphics, Machine Learning

Keywords : Virtual reality, Avatar, Embodiment

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[OPEN] [ISL_CG_CV_DM_VP_047] Transform coding for dynamic point clouds 

3D point clouds have emerged as an advanced 3D representation. Compression technologies are needed to reduce the amount of data required to represent a point cloud. As a result, MPEG has recently launched a Call for Proposals on Point Cloud Compression (PCC). Block-wise transform coding has been proven to be very successful in video coding. However, as the points of a point cloud distribute irregularly in 3D space and the 3D-blocks are very often not fully occupied, tools such as discrete cosine transform (DCT) cannot be directly applied to compress point cloud. In this internship, the aim is hence to exploit new transform coding methods for irregular point clouds which can achieve both good compression performance and low complexity. This internship proposal focuses on transform coding methods for the intra and inter coding mode modules of dynamic point clouds. Furthermore, the global transform coding methods for the intra frames of dynamic point clouds will also be studied

Skills : dynamic point clouds, 3D-block based prediction, transform coding, graph Fourier transform

Keywords : video coding, graph based signal processing, geometry processing

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[OPEN] [ISL_CV_DM_HCI_048] Deep learning for gesture recognition  

Gestures are a common form of human communication. Researchers and engineers have thus naturally developed techniques to use this form of communication to interact with computers. The development of new sensors, e.g., Kinect-like sensors, and new deep learning techniques have improved significantly the performance of gesture recognition systems. Nevertheless, these techniques are still demanding in terms of computational resources, and quantity of data for training. Improvements are still needed to make these systems work in complex environment. In this internship, the goal is to explore new deep learning techniques and the use of new type of data, e.g., event-based camera, to reduce computational costs and the quantity of data for training

Skills : A good candidate will have confidence with computer vision, deep learning and the implementation of algorithms using languages such as C/C++ or Python and libraries such as Tensorflow or Theano

Keywords : gesture recognition, hand pose estimation, deep learning

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[CLOSED] [HOME_DM_049] Artificial versus natural voice classification for vocal interface usage 

Vocal interfaces are now more and more popular, Amazon Echo, Google Home propose such convenient interfaces to facilitate the end user life. A keyphrase/keyword (e.g. “Alexa”, “OK Google”) pronounced by the end user is usually the trigger to access the vocal interface service. In the case the end user’s TV or audio system broadcasts the keyphrase accidently (cf. “OK Google” in SuperBowl advertisement : https://www.theverge.com/2017/2/5/14517314/google-home-super-bowl-ad-2017), the vocal interface takes actions without the end user control. The student shall explore different solutions to circumvent this problem in discriminating/classifying artificial against natural vocal sources. He will also build a prototype based preferably on a raspberry Pi platform to demonstrate the result of his internship

Skills : machine learning, classification, notion of audio processing, Scikit-learn, Python, C, MATLAB, linux (PC, Raspberry Pi)

Keywords : vocal interface, machine learning, acoustic, audio processing, features extraction

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[OPEN] [HOME_DM_050] Learning from real home audio 

With the advent of smart speakers and other microphone-equipped home agents there is the potential for using audio to monitor human activities, learn household habits and detect anomalies caused by, for example, unwanted intrusions, equipment failures or human incidents. However real home audio is unstructured and noisy so the automated inference of useful information is not straightforward. The internship will analyse a pre-existing audio dataset which was recorded in a family home. The objective is to be able to recognise household activities from the audio and to explore the feasibility of using machine learning techniques to characterise habits and detect anomalies

Skills : Audio Processing, Machine Learning, Programming

Keywords : Artificial Intelligence, Smart Home

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[OPEN] [HOME_DM_NW_052] Learning of in home device behaviour through experimental analysis 

In the context of Wi-Fi full home coverage solutions that are emerging on the market, multiple access points (AP) are installed in homes and user devices automatically select the best AP to use (typically the one with the strongest signal). However, this selection is not optimal under certain situations. The goal of the internship is to characterize Wi-Fi devices to improve an algorithm that enables to find the best configuration whatever the home environment. In a first step, roaming behaviors of mobile devices will be captured in our lab testbed to build a dataset. The idea is to collect in a controlled environment data related to operational Wi-Fi behavior of a large set of devices. In a second step, characterization of devices and their behavior will be conducted on the experimental data by using different technologies such as unsupervised learning (clustering, …), probabilistic modeling using mixture/topic models or hypothesis testing

Skills : Machine learning (Clustering, …), Notion of wireless networking, Scikit-learn, Python, Linux (Raspberry Pi)

Keywords : Machine Learning, Statistical Analysis, wireless whole home coverage (Wi-Fi), device behaviour characterization

 

This internship is located in Rennes, France. If interested, please apply at stage.technicolor@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.