Object recognition, detection and localization 101

[yugrocks]

• Abstract (2-3 lines): This talk will cover the state of the art techniques in object detection and localization, face recognition and some applications of deep learning in computer vision. This will be more of a case study based approach.

• Brief Description and Contents to be covered:

  • What is object detection, localization, recognition.
  • The state of the art techniques:
    • Object detection using Haar Cascades , Histogram of Oriented Gradients (hands-on with openCV2 in python)
    • Basics of Convolutional neural networks (CNNs) and different standard architectures: VGG16, Inception, Residual Network.
    • Object recognition using CNNs.
    • Sliding windows using CNN, its advantages and disadvantages.
    • Improved Deep Learning techniques namely Mask R-CNN, Fast-RCNN, Faster-RCNN, YOLO
    • The YOLO (You Only Look Once) algorithm for object localization and recognition in detail.
  • One shot learning and Face recognition (very brief).
  • A brief case study of the newly released software, Detectron (from Facebook AI research).

• Pre-requisites for the talk Basic knowledge of Machine Learning classifiers like SVM, Neural Networks and basics of python language.

• Time required for the talk 1 hour.

• Link to slides Soon

• Will you be doing hands-on demo as well? Yes.

• Link to ipython notebook (if any) Soon.

• About yourself Btech. CSE student at Maharaja Surajmal Institute of Technology, currently researching on information extraction and document classification in NLP at Grainpad Pvt. Ltd.

• Are you comfortable if the talk is recorded and uploaded to PyData Delhi's YouTube channel ? Yes

• Any query ? No

[shagunsodhani]

Hey, @yugrocks Thanks for the proposal. I think it would be interesting to see how HOG features perform against vanilla CNNs. You might want to consider splitting the talk into two sessions. The second session could be:

• Improved Deep Learning techniques namely Mask R-CNN, Fast-RCNN, Faster-RCNN, YOLO
• The YOLO (You Only Look Once) algorithm for object localization and recognition in detail.

[yugrocks]

Hi @shagunsodhani . Thanks for the suggestion! However, I am a little confused about what you mean bysplitting the talk into two sessions. Will you please explain? :)

[Dawny33]

@yugrocks He meant, 2 half-an-hour sessions. Half of the topics in one and other half in the other.

It doesn't have to be exactly half-and-half, just that the topics need to be coherent in both the sessions.

[yugrocks]

Okay. That sounds good. We can do that. :)

[manojpandey]

Hi @yugrocks - any update on slides? We're having the meetup this Saturday!

[yugrocks]

Hi @manojpandey I have decided to talk about the other proposal that I had submitted (about Deep Learning in NLP), for this week. So, I'm currently preparing slides for that one only.

[Ridhwanluthra]

@yugrocks any update on the slides? We are planning a meetup on 16th and would love you have your talk then.

[yugrocks]

Hi @Ridhwanluthra, my exams are ending on 13th June. I'll try to complete the slides prep by 14th.