Object Detection using deep convolutional network
Submitted by Koustubh Sinhal (@koustubh) on Tuesday, 31 May 2016
The talk aims to cover the advancements in object detection framework based on convolutional networks. The goal of object category detection is to identify and localize objects of a given type in an image. The main challenge associated with object detection in unconstrained images is to localize the object regardless of its location and scale. I will cover various state of the art methods employed in generic object detection starting from rcnn that employs a third party generic object proposals(selective search) to faster-rcnn which generates the proposals from the convolutional network itself with substantial reduction in running time. I will then cover the role of context for object detection in domains like fashion where region based methods fall short of predicting the correct category and methods to incorporate the context in existing object detection systems.
1. Object detection framework with convolutional network
2. Region based object detection algorithm (R-cnn)
3. Methods to speed up the training and testing time
4. Bypassing the generic object proposal phase(faster-rcnn)
5. Context(surrounding of region) Importance for object detection in fashion domain
6. Methods to incorporate context
Koustubh Sinhal is the founder and CTO of iLenze.com, an AI platform focussed on visual search. He completed his bachelor and Masters degree from IIT Kanpur. He has a long standing interest in Machine learning, Artificial intelligence and computer vision. His novel work at iLenze is currently patent pending and opens door for next generation AI which works for all kind of real-world-images with huge clutter and noise.