Chris Pollett > Students >
Matharu

    ( Print View)

    [Bio]

    [Blog]

    [CS 297 Proposal]

    [Deliverable 1]

    [Rapid Object Detection using a Boosted Cascade of Simple Features-PDF]

    [Deliverable 2]

    [Deliverable 3-PDF]

    [Deliverable 4]

    [CS297 Report-PDF]

    [CS298 Proposal]

    [CS298 Report-PDF]

    [CS298 Defense Slides-PDF]

Deliverable 1

Multi-Class Classification and Detection on Emoji Dataset

Goal: To perform multi-class Classification and Detection on self-generated emoji dataset using Keras and Open CV

Aim: To get acquainted with the concepts of machine learning and artificial intelligence

Conclusion: Performed emoji detection through OpenCV and classification through Keras. Low accuracy due to imbalance of images present under each emoji class.

This deliverable has been divided into 2 parts.

Deliverable 1.1: Object Detection

In this part, object detection is performed with the help of OpenCV. The number of emojis present in an image is detected with the help of CV2 Hough Circles.

Emoji Detection Code

Here are some of the results:

Detecting single emoji in an image
Detection 1

Detecting two emojis in an image
Detection 2

Deliverable 1.2: Object Classification

In this part, we classify all the emojis into best predicted classes present in an image. Here all the images will contain upto 3 emojis. To achieve this a simple neural net is designed using Keras to perform multi-class classification. The accuracy of the model is not at its best level due to some inconsistencies in the dataset.

Emoji Classification Code

Model Architecture
Model Arc

Here are some of the results:

Correct Classification(Predicted Classes)
Classification 1

Incorrect Classification(Predicted Classes)
Classification 2

The accuracy of the model is poor due to inconsistent data present under each class.
accuracy

Dataset

Emojis dataset was generated using PIL library. 5 kinds of emotions were chosen and upto 3 of them were written to the image. Random emojis size were placed randomly in any position in the image. A total of 1000 such images were created. This dataset was further divided into training and test dataset.

Generation of Dataset Code

Image consisting of 1 emoji
1 Emoji

Image consisting of 2 emojis
2 Emojis
Image consisting of 3 emojis
3 Emoji