The mission mainly consists in improving the accuracy of the algorithm we are developing on a proprietary video dataset. Tasks will include:
-Tracking down and implementing state-of-the-art video recognition deep learning architectures [1-2-3-4]
-Implementing a light optical flow based action recognition model 
-Designing intermediary descriptors [4-5]
-Designing an unsupervised approach to the problem 
-Mastering general data preprocessing techniques  and temporal data augmentation
-Implementing an efficient pruning for all neural networks used
-Improving the multithreading and multiprocessing training and inference pipelines
-Collaborating with the Integration team and the Product Development team
-Assisting the CTO in tech related appointments and collaboration with labs
-Explore new tasks related to the algorithm arising from tech watch
– Proven expertise in Deep Learning, especially in computer vision (video related tasks are a plus)
– Interest in research and ability to extract useful information from scientific papers
– Solid grasp of at least one of the following deep learning frameworks : Pytorch, Tensorflow, CNTK, Caffe
– Understanding of convolution and famous related architectures (resnext, I3D, RPN, two-stream networks)
– Analytical mind, ability to take a step back and see the big picture
– Problem-solving aptitude
– Master / PhD in Deep Learning or relevant fields.
 Bolei Zhou, Alex Andonian, Antonio Torralba : Temporal Relational Reasoning in Videos arXiv:1711.08496
 Jiawei He, Mostafa S. Ibrahim, Zhiwei Deng, Greg Mori : Generic Tubelet Proposals for Action Localization arXiv:1705.10861
 Sijie Yan, Yuanjun Xiong, Dahua Lin : Spatial Temporal Graph Convolutional Networks for Skeleton-Based Action Recognition arXiv:1801.07455
 Max Jaderberg, Karen Simonyan, Andrew Zisserman, Koray Kavukcuoglu : Spatial Transformer Networks arXiv:1506.02025
 Zhenzhong Lan, Yi Zhu, Alexander G. Hauptmann : Deep Local Video Feature for Action Recognition arXiv:1701.07368
 Eddy Ilg, Nikolaus Mayer, Tonmoy Saikia, Margret Keuper, Alexey Dosovitskiy, Thomas Brox : FlowNet 2.0: Evolution of Optical Flow Estimation with Deep Networks arXiv:1612.01925
 Thomas Schlegl, Philipp Seeböck, Sebastian M. Waldstein, Ursula Schmidt-Erfurth, Georg Langs : Unsupervised Anomaly Detection with Generative Adversarial Networks to Guide Marker Discovery arXiv:1703.05921
Paris – Full time
€40k – €80k
Founded by Polytechnique and HEC alumni, Veesion is incubated at Agoranov (96 bis Boulevard Raspail, 75006 Paris).
We are developing a gesture recognition technology in video content. The potential applications are diverse and our first focus is shoplifting detection in the retail industry.
We built a large video database thanks to partnerships with major retailers and we are in constant discussion with top-notch Computer Vision labs such as the Thoth and Willow teams in order to drive our research.
We are a young and agile startup: joining us now is the opportunity to grow with the company.