Dr. Amit Kale
Data management for AI
„In science, you can say things that seem crazy, but in the long run they can turn out to be right. We can get really good evidence, and in the end the community will come around.“ - Geoffrey Hinton
I am a principal senior expert in computer vision at the Research and Technology Center in India. My research is motivated by the desire to organize and manage large scale video and multi-sensor data (Peta Bytes) with the goals of being able to smartly curate the right data desired by algorithm development. This includes automated approaches to select the most representative sub set of a large set of images, and search and retrieve scenes of interest from the stored images. Our research has multiple goals such as reducing the cost of ground truth generation by removing redundancies, supporting function development and testing to find difficult cases where algorithms do not work well, which can then be used to collect more of such cases or synthetically generate them. In order to achieve this we explore the structure and representation power of deep convolutional neural networks. We develop human computer interfaces that go hand in hand with the deep learning approaches to ensure ease of usage by the end users.
- Head of research group, imaging and computer vision, Siemens Healthcare
- Assistant research professor, University of Kentucky
- Doctoral Degree, University of Maryland College Park
Amit Kale (2004)Identification of Humans Using Gait
- Amit Kale, A. N Rajagopalan, A. Sundaresan, N. Cuntoor, A. RoyChowdhury, V Kruger, Rama Chellappa
- IEEE Transactions on Image Processing September 2004
- DOI: 10.1109/TIP.2004.832865
Samarjit Das (2012)Particle filter with Mode Tracker for Visual tracking across Illumination Changes
- Samarjit Das, Amit Kale, Namrata Vaswani
- IEEE Trans. on Image Processing
Venkata Gopal Edupuganti (2018)Automatic Optic Disk and cup segmentation of fundus images using deep learning
- Venkata Gopal Edupuganti, Akshay Chawla, Amit Kale
- ICIP 2018
- DOI: 10.1109/ICIP.2018.8451753
Amit Kale (2006)A Joint Model of Illumination and Shape for Visual tracking
- Amit Kale and Christopher Jaynes
- Proceedings of IEEE Computer Society Conference on Computer Vision Pattern Recognition 2006
- DOI: 10.1109/CVPR.2006.30
Etienne Grossmann (2005)Towards Interactive Generation of ”Ground Truth” in Background Subtraction from Partially Labeled Examples
- Etienne Grossmann, Amit Kale, Christopher Jaynes
- Proceedings of the IEEE Workshop on Visual Surveillance and Performance Evaluation of Tracking and Surveillance, October 2005 Beijing China
- DOI: 10.1109/VSPETS.2005.1570932
Rahul Thota (2016)Fast 3D Salient Region Detection in Medical Images using GPUs
- Rahul Thota, Sharan Vaswani, N. Vydyanathan and Amit Kale
- Machine Intelligence and Signal Processing
- Springer Verlag 2016
- DOI: 10.1007/978-81-322-2625-3_2
Danny Crasto (2005)The Smart Bookshelf: A study of camera projector scene augmentation of an everyday environment
- Danny Crasto, Amit Kale and Christopher Jaynes
- Proceedings of the IEEE Workshop on Applications of Computer Vision, Colorado Springs CO January 2005
- DOI: 10.1109/ACVMOT.2005.116
Bharat Sharma (2009)Towards a Robust, Real-time Face Processing System using CUDA-enabled GPUs
- Bharat Sharma, Naga Vydyanathan Rahul Thota, Amit Kale
- Proceedings of the HiPC 2009 Kochi, India
- DOI: 10.1109/HIPC.2009.5433189
Srikanth Cherla (2008)Towards Fast View Invariant Human Action Recognition
- Srikanth Cherla, Kaustubh Kulkarni, Amit Kale and V. Ramasubramanian
- Proceedings of the Ist Workshop on Computer Vision and Pattern Recognition for Human Computer Communicative Behavior held in conjunction with CVPR 2008 Anchorage Alaska
- DOI: 10.1109/CVPRW.2008.4563179
Kaustubh Kulkarni (2010)An unsupervised framework for action recognition using Actemes
- Kaustubh Kulkarni, Edmund Boyer, Radu Horaud and Amit Kale
- Proceedings of the ACCV 2010, Queenstown NZ
- DOI: 10.1007/978-3-642-19282-1_47
Interview with Dr. Amit Kale
Principal senior expert and group manager
Please tell us what fascinates you most about research.
There is beauty in the scientific method. Identifying a problem, applying the state of the art to it, discovering where it fails and thinking hard about how to overcome it, is an enriching activity. When you are focusing on a problem and a serendipitous discovery happens that can drastically change the storyline and lead to unexpected gains, this is particularly fascinating to me.
What makes research done at Bosch so special?
I find Bosch Research to be a very exciting place to be. People are deeply knowledgeable and passionate about the technology, which is why Bosch is the leading mobility solution provider. There is a wealth of data available from a huge variety of sensors which is one of the very basic requirements for building great AI systems. Furthermore, the people are very open and collaborative. With these three ingredients of intellectual curiosity, availability of data and openness, Bosch Research is a very attractive place for a researcher.
What research topics are you currently working on at Bosch?
Currently I am working in the area of data management for AI. The work spans methods of smart data curation for various tasks such as object detection, semantic segmentation etc., smart data selection approaches using active learning to identify hard examples, algorithms for flexible search and retrieval of images using various sources of data including image information, meta data from complementary sensors etc. In addition to this algorithm development, I also oversee the development of user friendly human computer interfaces, in consultation with the users. Being based out of India, we also bring our AI expertise to the biggest offshore development center outside Germany in Bangalore, supporting Bosch's regional initiatives in India.
What are the biggest scientific challenges in your field of research?
AI algorithms have to perform with extreme reliability and performance guarantees in an “open context” environment.This implies that the AI algorithms in these products have to be trained upfront with data that contains even the rarest situation that the products might encounter in the field. This entails the collection of an enormous amount of data with the associated challenges of storage costs and the challenge of sifting through them to find the right data to label.
How do the results of your research become part of solutions "Invented for life"?
The work done by my team in the area of smart data selection helps in the development of autonomous system development such as self-driving cars or security systems making them more intelligent and affordable. Our work in search and retrieval facilitates users to search, retrieve and download interesting and difficult images for their algorithms, making autonomous driving safer.