[CLOSED] [MCL_2016_CV_004] Image and video processing with an extended pixel representation

Light-Fields (LF) are gaining a lot of popularity in the field of computational photography because of the additional information they have compared to traditional imaging. Indeed, LF allow novel post-capture processing capabilities such as depth estimation or refocusing.

LF can be captured either with plenoptic cameras or an array of cameras. When representing LF, both position and direction information are available, extracted from data captured by cameras (a 4D dataset). LF can thus be represented by new elementary objects, extending the notion of pixel.

The goal of the internship is to implement LF processing or visual effect operations as refocus, viewpoint change or LF mixing using extended pixels as elementary bricks.

Skills : Major in computer vision or image processing. Optics is a plus

Keywords : light-field, computational photography

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

[CLOSED] [MCL_2016_CV_005] Light-Field Superresolution

Light-Fields (LF) are gaining a lot of popularity in the field of computational photography because of the additional information they have compared to traditional imaging. Indeed, LF allow novel post-capture processing capabilities such as depth estimation or refocusing.

LF can be captured either with plenoptic cameras or an array of cameras. However, the additional LF information comes at the cost of a resolution loss (angular or spatial resolution) which is the major bottleneck of the LF technology.

The goal of the internship is to design and implement a superresolution algorithm for LF images acquired with a plenoptic camera or an array of cameras.

Skills: major in computer vision or image processing

Keywords: light-field, computational photography, super-resolution

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

 

[CLOSED] [MCL_2016_CV_006] New method for calibrating Light Field cameras

Light-Fields (LF) are gaining a lot of popularity in the field of computational photography because of the additional information they have compared to traditional imaging. Indeed, LF allow novel post-capture processing capabilities such as depth estimation or refocusing.

LF can be captured either with single focal or multifocal plenoptic cameras or an array of cameras for example. The way the object space is sampled by the pixel array through the optical system characterize these cameras. This characterization plays a key role in the quality of future processing of the light field signal.

The goal of the internship is to contribute to the development of a LF Calibration platform and method able to evaluate with precision the acquired light field geometrical structure for various types of cameras (plenoptic / arrays…).

Skills: major in computer vision or image processing. Optics is a plus.

Keywords: light-field, computational photography

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

 

[CLOSED] [MCL_2016_CV_007] Plenoptic lens design optimization strategy

Plenoptic cameras are using a micro-lens array added into a conventional camera in front of the sensor. The sensor records a collection of micro-images which can be combined to compute various re-focused images with freely selected focalization distances. Such cameras have been recently popularized by the Lytro company.

By combining the micro-images into a single refocused image it is also possible to correct numerically some of the optical aberrations of the main-lens. Optimizing the optical design of the main-lens should be computed taking into consideration the optical aberrations which are corrected numerically when combining the micro-images. A new scheme of optical design should be defined specifically for plenoptic cameras.

The objective of the internship is to define an optimization strategy to improve the main-lens design. Reaching good compromise in optical design & numerical aberration correction, plenoptic super-resolution & demosaicing algorithms will be tested and benchmarked. This study will be performed mainly with ZEMAX software and synthetic images simulating real-lenses (possibly, the plenoptic Lytro Illum and Raytrix camera could be used).

Skills : Optical Design, Image processing, Computer Science.

Keywords : Light-Field – optical aberrations – re-focusing.

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

 

[CLOSED] [MCL_2016_CV_CG_HCI_VP_008] Dynamic and virtual media screen in domestic environment

The goal of the internship is to develop an Augmented Reality application for mobile device (typically a tablet), which allows a user to easily define and activate a virtual media screen attached to a wall of his choice, with personalized size, inside its house.

The internship will focus on natural image features detection & tracking in that particular context, using related computer vision libraries. The development will be typically done on Android or Windows tablet using related SDK and additional libraries that may help for the internship’s purpose.

Skills : Specific knowledge in computer vision, good programming skills in C++ , Android or Windows, possibly CUDA/OpenCL, 3D rendering library (Unity 3D, Unreal Engine)

Keywords : Computer vision, video processing, image feature detection, pose camera tracking, real-time processing.

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

 

[CLOSED] [ISL_2016_CV_VP_019] Light Field Fourier slice re-focusing

Light Fields are defined as all the light rays at every point in space travelling in every direction. It is essentially 4D data (position + direction). Light Fields can be captured either with plenoptic cameras or an array of cameras. Light Fields are gaining a lot of popularity in the field of computational photography because of the additional information they have compared to traditional (2D) imaging. Indeed, Light Fields allow novel post-processing capabilities such as refocusing.

The goal of this internship is to investigate the 4D Fourier spectrum of Light Fields and to implement re-focusing algorithms based on the extraction of 2D slices from the 4D Fourier domain.

Skills : Major in signal/image processing or computer vision. Skills in C++ programming or Matlab.

Keywords : Light Fields, image processing, Fourier transform, spectral analysis

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

 

[CLOSED] [ISL_2016_CV_VP_020] Perceptual coding for new gen video codec

New generation video codecs, such as H.265/HEVC or VP9, achieve high compression performances w.r.t. the PSNR metric. Perceptual approaches enable to improve even further those performances, while relying on quality metrics that consider the human visual system properties. This internship aims at optimising the use of such metrics within a compression framework.

Skills : video compression, image processing, machine learning, C/C++ programming

Keywords : new-gen video compression

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

 

[CLOSED] [ISL_2016_CV_VP_021] Hi Mom! Video chat-specific video compression algorithms

Current commercial video-chat solutions do not take advantage of signal redundancies intrinsic to the application. Despite the fact that these systems operate in closed-world situations, they use standard codecs that, given the bandwidth constrained nature of the task, result in dropped frames, blocking artifacts and a poor user experience. Yet the closed-world nature of video-chat systems means that the codec at both ends of the communication link does not need to adhere to standardized options optimized for generic video content.

In this internship, the aim is hence to exploit the high redundancy that video-chat signals benefit from in developing a codec optimized for the application. The tools envisioned to carry out this include machine learning, overcomplete transforms, sparse coding, online optimization methods and information theory.

Keywords: Video compression, sparse coding, machine learning, face processing

A good candidate will hence need to have confidence with mathematics and implementing mathematical algorithms using languages such as Python, Matlab, or C/C++. Furthermore, we envision putting together a functional demonstration model, and hence a plus would include having some notion of interface development and web sockets, preferably using Python.

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

[CLOSED] [MCL_2016_CG_CV_HF_027] Study of cinematographic effects for fully immersive movies

The goal of the internship is to imagine, develop and prototype cinematographic effects for full immersive movie experience; those effects should enable the director to guide the user’s attention and maximize the visual comfort. Concepts will then be validated by a user test panel.

Skills: Specific knowledge in 3D Engine (Unity/C# or Unreal/C++ engines), and shader implementation. Interest in cinematography, storytelling and in user perception.

Keywords: Computer vision, computer graphics, HMD, cinematography, user perception.

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

 

[CLOSED] [ISL_2016_CV_VP_028] Multimodal-based Interestingness Prediction

Knowing whether a media content, i.e., image or video, is interesting for a viewer has numerous applications from assets management, improved education, to targeted advertising.

This internship proposal targets the development and implementation of such an interestingness prediction algorithm based on machine learning techniques. One expected output is to submit a system to the 2016 MediaEval (http://www.mutimediaeval.org) task on interestingness.

Skills: machine learning, computer vision, image processing, C++, Python, OpenCV. Audio processing skill would be a plus.

Keywords: multimodal processing, machine learning (deep learning), image video interestingness.

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

 

[OPEN] [ISL_2016_CV_033] Automatic detection of tracked object with handheld cameras

The goal of this internship is to analyse, design and implement a full-fledged solution to the problem of automatically determining the object being tracked on-live by a user with a consumer camera. The intern will combine computer vision theory and algorithms along with user interaction tools in order to design a rich user-centric feature to be integrated into new video devices to come.

Skills: Engineer or master student with a solid background in computer vision and image/signal processing. Experience with motion analysis is a plus.

Strong mathematical background.

Good programming (C++, python, Matlab) and software design skills.

Good written/oral communication skills.

Fluent in English

Keywords: Computer vision, object tracking, motion analysis, motion/image saliency, user interfaces.

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

 

[CLOSED] [ISL_2016_CV_VP_037] Deep Visual Search

The goal of this internship is to analyse, design and implement a full-fledged solution to the problem of automatically determining the object being tracked on-live by a user with a consumer camera. The intern will combine computer vision theory and algorithms along with user interaction tools in order to design a rich user-centric feature to be integrated into new video devices to come.

Skills: Engineer or master student with a solid background in computer vision and image/signal processing. Experience with motion analysis is a plus.

Strong mathematical background.

Good programming (C++, python, Matlab) and software design skills.

Good written/oral communication skills.

Fluent in English

Keywords: Computer vision, object tracking, motion analysis, motion/image saliency, user interfaces.

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

 

[CLOSED] [ISL_2016_CV_VP_039] Multi-image editing

Motivated by exponentially growing collections of personal photos that are stored and shared on-line, recent works investigate various ways of enhancing photos collectively. The rationale is two-fold: related pictures (eg from the same event) should be enhanced in a consistent way; low quality pictures should benefit at most from the content of related high quality ones. The goal of this internship is to devise and experiment novel ways to perform such a multi-enhancement, with a specific focus on the problem of color consistency across different views of the same scene. 

Keywords: image enhancement, color homogenization, multi-image manipulation.

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

 

[CLOSED] [MCL_2016_CG_CV_VP_048] Real-time 3D surface reconstruction and tracking

The objective is to develop a real-time application that can build a 3D scene model from data captured by a sensor moving around the scene.

The main components are depth measurement, pose estimation and incremental reconstruction

The internship will consist of an analysis of the state of the art, the specification of a solution suited to our needs, and the implementation, testing and improvement of a method.

Skills : Specific knowledge in computer vision, good programming skills in C/C++ , Windows, possibly OpenGL ES/ CUDA.

Keywords: Computer vision, 3D scene analysis, video processing, 3D reconstruction, pose estimation, real-time processing.

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

 

[CLOSED] [ISL_2016_CV_DM_051] Joint Modelling of Image Representation and Description using Deep Learning for categorization

Summary - The extraction of semantic metadata from images and video is a challenging task that has many applications ranging from video classification and indexing for multimedia retrieval systems to photos organization in mobiles phones. The advent of numerous and large training sets (eg Imagenet, Places) and the maturity of recent deep learning software libraries (eg Theano, TensorFlow, Caffe ) make new training architecture possible that can potentially deliver better precision or finer categorization/understanding.

The topics of the internship is to implement emerging approaches to image classification relying on deep learning using *simultaneously* several training datasets pertaining to different modalities : categorized images dataset, images description, pure textual/logic corpora.
-The primary goal is to achieve better precision on image categorization (to be evaluated on academic challenges).
-To be able to perform efficient categorization on small specific dataset.

Motivation - The size of the models learnt through deep learning technique (the number of coefficients) is so large that specific and ad-hoc tricks were required to prevent overfitting. The motivation behind the proposal is to find a way to help the model to converge faster using auxiliary datasets that will be used jointly (enforcing sharing parts of the model)

Goal The research work to is to propose and build an effective deep architecture for image categorization to be trained on datasets to be identified and then to conduct evaluation on public academic benchmarks. A multi-GPU platform will be available for calculation.

Keywords image classification, deep learning, joint reasoning.

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

 

[OPEN] [MCL_2016_CG_CV_056] Depth estimation from disparity using hybrid camera array

With the rise of VR devices such as Oculus Rift, Samsung Gear or HTC Vive, more and more immersive contents are available for consumers. Such content is generally obtained with computer graphic animation. New ways of capturing real scenes to create immersive content now compete with traditional CGI, for instance using camera arrays (e.g. Google Jump). For those scenes to be immersive, a good depth estimation of the elements composing the scene is needed.

This internship targets the study, implementation and performance analysis of a depth estimation solution based on a hybrid camera array (vision and IR).

Skills : C++/Python/OpenCV, fluent in English, good written and oral communication skills.

Keywords : depth estimation, disparity, stereo vision,  camera array, calibration, matching.

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