Skip to content

Latest commit

 

History

History
61 lines (34 loc) · 3.66 KB

README.md

File metadata and controls

61 lines (34 loc) · 3.66 KB

Problem Statement - Task entails you to develop a simple Web Application, which when given an image, is able to detect a face (show with a boundary box) and give a result saying whether Narendra Modi and/or Arvind Kejriwal are present in the image or not.

TL;Dr- The Web Application is up at http://image-recog.eigencoders.com / http://104.196.154.231 (Accuracy 94.7%):smiley::smiley:

First thought, this looks like a Image Classification Task, Machine Learning approach would do.

So, I broke down the problem into parts -

  • Collect Images of classes namely Narendra_modi, Arvind_kejriwal, peoples
  • Extract Face from Images of different classes
  • Divide Data into three categories Training, Test, Cross validation
  • Train the Classifier on Images from each Class on their Training set.
  • Test it on Test_Set

The files are accordingly as -

The best output was given by Logistic_Regression_Classifier, Accuracy (94.7%)

I saved the Classifier as .pickle object named Logistic_regression_fit.pickle

So now, whenever a new image comes to me, I have to do following processes on it

  • Extract faces, crop, align and create feature numpy array
  • load my Object for Logistic regression
  • Run the feature array on classifier, get label and accordingly show output.

Test it now, 💯 http://image-recog.eigencoders.com / http://104.196.154.231 😎

Things which were planned but could not be done 😞.

  • Wanted to save the datset of Train, Test and Cross Validation data in MongoDB, but it turns out that maximum size of BSON document size is 16 MB which can be stored in mongo document. It is visible in in Block [71] of Image_Classification.ipynb where it throws an error DocumentTooLarge: BSON document too large (1484657743 bytes) - the connected server supports BSON document sizes up to 16793598 bytes(16 MB)😕. So I worked around it by saving the datset in .pickle format, which allows me use dataset directly without computating again and again.
  • Wanted to upload dataset_train_final.pickle on this git repo, but can't due to size restriction on github. I plan to host all dataset_train_final.pickle, dataset_test_final.pickle and dataset_cv_final.pickle on google drive and provide link.

Web Deployment Part

I deployed the app using Flask Library of python on my Google Cloud account. Link to Github Repo

Reference

I would like to thank several tutorials on machine learning, Reddit.com/r/MachineLearning has awesome learning resources 🙇, also OpenFace for making Face detection Library and developers of Flask for making deployment very easy.

Sample of Face Recogniser

Link - https://imgur.com/a/u4bgX Sample

For further developement/queries of this work, you can write me to : vnds.20150389[@]btech[.]nitdgp[.]ac[.]in :bowtie:

If you are impressed by this, then you should check out my other works on my personal page here