Saumya Jetley

TAU Team, INRIA-Paris Saclay       saumya.jetley@inria.fr

I am a postdoc with Michele Sebag at the TAU Team, INRIA-Paris Saclay. I work on causal inference in the context of French job market, alongside the industrial partner QAPA, and on privacy-preserving predictive analysis for French covid data. I completed my PhD in computer vision and machine learning algorithms for AI systems, with a special focus on interpretability and robustness, under Phil Torr at Oxford University.

My research work has been variously funded by the Government of India, European research council, US department of defense and Five AI (autonomous driving startup). I am an IET Postgraduate Research award holder for 2019, have published several papers in top-tier CV and ML conferences, was selected for the FDL NASA programme 2019, and was an invited panel member at FDL Europe Big Think 2020. I have presented my research at various international conferences (NeurIPS, CVPR, ICLR) and workshops (Women in Machine Learning, Deep Learning Edinburgh, Reconfigured Vision), and am grateful to have been an invited speaker at institutes such as AI Wadhwani, ISI Kolkata, IIT Kharagpur, INRIA Paris Central, INRIA Paris Saclay.


News

10/19: I am joining TAU Team as a postdoc on causal learning and inference. *Book of Why* has opened new doorways!
08/19: I have spent this summer visiting ISI, Kolkata and IIT, Kharagpur as an invited guest.
06/19: I gave an invited talk at the Willow team - INRIA-Paris Central.
04/19: I successfully defended my thesis to an examining panel of Prof. Andrew Zisserman and Prof. Tinne Tuytelaars.
03/19: I was invited for a talk at the TAU team - INRIA-Paris Saclay, details.
02/19: I will be a visiting researcher at FiveAI starting this month.
01/19: I submitted my thesis, wohoo! 'Use and Examination of CNNs for Scene Understanding'.
12/18: I gave a talk at the newly inaugurated Wadhwani Institute, Mumbai. Such energy, great leadership, AI for social good in Indian space, ftw!
10/18: Excited to be giving a talk on my academic journey and work at an event by EF and Researc/hers code.
10/18: Grateful to be receiving the WiML travel award, looking forward to the workshop.
10/18: Grateful to be receiving the NIPS travel award.


Older news



Publications

[2019]
Straight to Shapes++: Real-time Instance Segmentation Made More Accurate, Laurynas Miksys, Saumya Jetley, Michael Sapienza, Stuart Golodetz, Philip H.S. Torr, Arxiv preprint 2019 [pdf] [code] [bibtex]

[2018]
With Friends Like These, Who Needs Adversaries?, Saumya Jetley*, Nicholas A. Lord*, Philip H.S. Torr, Proceedings of the 32nd conference on Neural Information Processing Systems (NIPS) 2018 [pdf] [code] [poster] [bibtex]

Learn to pay attention, Saumya Jetley, Nicholas A. Lord, Namhoon Lee, Philip H.S. Torr, Proceedings of the 6th International conference on learning representations (ICLR) 2018 [pdf] [code] [bibtex]

[2017]
End-to-end Saliency Mapping via Probability Distribution Prediction, Saumya Jetley, Naila Murray, Eleonora Vig, U.S. Patent 9,830,529 with Xerox Corp (now Naver Labs) [details]

Straight to Shapes: Real-time Detection of Encoded Shapes, Saumya Jetley*, Michael Sapienza*, Stuart Golodetz, Philip H.S. Torr, Proceedings of the International conference on Computer Vision and Pattern Recognition (CVPR) 2017 [pdf] [code] [bibtex]

[2016]
End-to-End Saliency Mapping via Probability Distribution Prediction, Saumya Jetley, Naila Murray, Eleonora Vig, Proceedings of the International conference on Computer Vision and Pattern Recognition (CVPR) 2016 [pdf] [code] [bibtex]

[2015]
Prototypical Priors: From Improving Classification to Zero-Shot Learning, Saumya Jetley, Bernardino Romera-Paredes, Sadeep Jayasumana, Philip H.S. Torr, Proceedings of the British Machine Vision Conference (BMVC) 2015 [pdf] [code] [bibtex]

[2014]
3D Activity Recognition Using Motion History and Binary Shape Templates, Saumya Jetley, Fabio Cuzzolin, Workshop proceedings of the Asian Conference on Computer Vision (ACCV) 2014 [pdf] [bibtex]

Multi-script Identification from Printed Words, Saumya Jetley, Kapil Mehrotra, Atish Vaze, Swapnil Belhe, proceedings of the International conference on Image Analysis and Recognition (ICIAR) 2014 [pdf] [bibtex]

[2013]
Automatic flag recognition using texture based color analysis and gradient features , Saumya Jetley, Atish Vaze, Swapnil Belhe, Proceedings of the International conference on Image Information Processing (ICIIP) 2013 [pdf] [bibtex]

Unconstrained handwritten Devanagari character recognition using convolutional neural networks, Kapil Mehrotra, Saumya Jetley, Akash Deshmukh, Swapnil Belhe, Proceedings of the 4th International workshop on Multilingual OCR (MOCR) 2013 [pdf] [bibtex]

[2012]
Hindi handwritten word recognition using HMM and symbol tree, Swapnil Belhe, Chetan Paulzagade, Akash Deshmukh, Saumya Jetley, Kapil Mehrotra, Proceeding of the workshop on Document Analysis and Recognition (DAR) 2012 [pdf] [bibtex]

Two-Stage hybrid binarization around fringe map based text line segmentation for document images, Saumya Jetley, Swapnil Belhe, V.K. Koppula, Atul Negi, Proceedings of the 21st International Conference on Pattern Recognition (ICPR) 2012 [pdf] [bibtex]


Datasets

Long long time ago, I gathered a dataset of 'Country Flags in the Wild'. It contains 12,854 train images and 6,110 test images of the flags of 224 different countries harvested from the world wide web and manually cropped to loosely fit to the inlying flags. More details can be found in this paper. To get access to the dataset, please get in touch with me by email. The demo of our android app using an ML model trained on this dataset can found here.


Talks

08/19: Visiting speaker at ISI, Kolkata and IIT, Kharagpur.
06/19: Invited speaker at Willow team - INRIA-Paris Central.
06/19: Presentation to the IET awards panel. Slides and accompanying video.
03/19: Invited speaker at TAU team - INRIA-Paris Saclay, details.
12/18: Invited talk on robustness properties of classification CNNs at AI Wadhwani Institute, Mumbai. Slides here.
10/18: Invited talk on my research work and academic journey at an event jointly organised by EF and Researc/hers code.
07/17: Invited talk on real-time instance segmentation at the Cambridge office of FiveAI. Slides here, demo here.
06/17: Demo'ed our instance segmentation algorithm at the Reconfigured Vision Workshop in London.
03/17: Invited talk on real-time instance segmentation at the Deep Learning Workshop'17 in Edinburgh. Recording, slides, and demo.


Teaching

I like teaching and have tutored for the following courses at the department of Engineering Science during my PhD:
- Attention in AI (Special exchange programme, Designed and taught the 10-hour course alongside Nick Lord, Michaelmas'18)
- B14 course on Image Analysis (Preparatory course, Trinity'17)
- B14 course on Signal Analysis (Hilary'17)
- B14 course on Image Analysis (Michaelmas'16)
- B16 course on Operating Systems (Hilary'16)

I have also been a laboratory demonstrator for the courses below:
- B14 Lab course on Image Analysis (Hilary'17)
- P5 Matlab software programming course (Michaelmas'16, Hilary'17, Trinity'17)


Reviewer duties


- I will be on the program committee for IJCAI'21.
- I have served as a reviewer for NeurIPS'18, ECCV'20, NeurIPS'20.
- I served on the program committee of CVPR 2020 Workshop on Adversarial Machine Learning in Computer Vision.
- I co-organised the Frontiers of Computer Vision workshop at the Deep Learning Indaba'18.


Outreach

As a member of several initiatives to encourage STEM education, I have frequently conducted coding sessions and educational talks for school students at science festivals and organised learning events for families and adults in science museums. Over the years, I have enjoyed pursuing scientific ideas for impact design and value discovery for different communities, and continue to do so.

09/18: I presented my work on realtime instance segmentation at the CS dept. open-day to prospective students and their parents.
06/18: I will be the Outreach & Scholarship officer for OxWoCS for the academic year 2018-19.
06/18: I co-organised a coding masterclass for girls aged 9-12 using spheros as part of the InspireHer! series.
05/18: I co-organised a coding workshop for students aged 12-13 at the Hay Science Festival.
06/17: I will be the Industry coordinator and External relations officer with OxWoCS for the academic year 2017-18.
05/17 to 06/17: As a student outreach ambassador for the Department of Engineering Science, University of Oxford, I have engaged with yr. 11-13 students at a host of events including - Lubbock lectures for schools, International Women in Engineering Day and Engineering Open Days. One set of slides especially popular amongst students during my classroom talks can be found here.
06/17: I gave a talk under the theme of 'Mobile Robotics' for the Looking Forward initiative of the Department of Computer Science, University of Oxford to encourage and motivate girls aged 14-15 towards STEM.
02/17 - 05/17: I served as a student volunteer with the Robots exhibition at the Science Museum London, supported by the Royal Academy of Engineering. More details here.


Awards

2019: Winner of the IET Postgraduate research award.
2017: Winner of the Tri-innovate challenge - Oxford university business innovation challenge.
2016: Amongst the top-10 algorithms for breast cancer prognosis challenge (lookout for Team HERcules).
2015: Best internship presentation award at Xerox Research Centre Europe, Grenoble (now Naver Labs).
2015: Honorary Mention for Originality for the Technical Essay entry at the International Computer Vision Summer School'15.
2014-15: Recipient of the Sir Richard Stapley academic scholarship.
2012: Best Performer– 2012 & Best Team – 2012 awardee at CDAC-Pune for outstanding research work.
2010: Winner of tech innovation award at Society for Computer Science Technology and Research's tech symposium
2006: All India Rank 102 in 5th National Cyber Olympiad of the Science Olympiad Foundation.