Project Blog

Nov 4, 2019

  • Performed several experiments with and without context.
  • Quantified the results. Results seem to be better with context which helps by giving the context about what the sentence is about.
  • Propose some use cases.

Oct 22, 2019

  • Trained the model on increased dataset. Tuned the model to improve the accuracy

Oct 8, 2019

  • Developed the model and trained on small sample dataset.

Sept 24, 2019

  • Developed the flow to extract text and context from images using Google APIs

Sept 10, 2019

  • Analyzed the sample dataset collected. Continue collecting the dataset throughout the semester.
  • Discussed the implementation techniques
    • Model 1 : Use sequence translation model to translate the text in the image.
    • Model 2 : Use model to summarize the images.
    • For the ambiguous translations in model1 use results from model2 to find the translation with the highest probability.

Sept 3, 2019

  • Discussed more about dataset collection for the project.

Aug 27, 2019

  • Group meeting to decide timings and discuss CS 298 proposal.

May 6, 2019

  • Submitted the project report

Apr 30, 2019

  • Cancelled the meeting

Apr 23, 2019

  • Discuss Deliverable 4 to extract the dataset

Apr 16, 2019

  • Present the paper related to text detection in images
  • Discuss Deliverable 3
  • Discuss Deliverable 4 to extract the dataset

Apr 9, 2019

  • Discussed text extraction techniques
  • Discuss Deliverable 3
  • Read papers related to text detection/extraction

Mar 26, 2019

  • Discussed techniques for image translation (xgettext, .po, .mo files)

Mar 19, 2019

  • Discussed the second deliverable
  • Finalize the deliverable 2 after doing minor changes
  • Read some more papers related to language translation

Mar 12, 2019

  • Present the paper on Language translation
  • Start text detection from images using openCV libraries

Mar 4, 2019

  • Discuss deliverable_2
  • Find dataset and start text detection using openCV libraries
  • Read a paper on language translation

Feb 26, 2019

  • Downloaded images from google to improve dataset. CNN model gave better accuracy for this new dataset
  • Improve the dataset for deliverable 1 and do some more experiments
  • Update the document for Deliverable 1
  • Find relevant papers related to the project and update the document

Feb 19, 2019

  • Discussed the deliverable_1 that was developed using both linear regression and CNN and building of dataset with openCV
  • Discussed strategies to improve the accuracy for CNN implementation
  • Increase the size of dataset
  • Print activations at each stage for filters applied
  • Print the various filters being applied

Feb 12, 2019

  • Decided on sub projects for understanding and building the main project
  • Plotted out more concrete deliverables required for 297.
  • Start working on deliverable_1 to detect suit and values on playing cards using CNN.
  • Decide the platform and libraries for developing the project and also for image transformation for the dataset.
  • Develop first cut of this project.

Feb 5, 2019

  • Discussed the project in detail
  • Discussed about neural networks in general
TODO: Make necessary changes to 297 proposal

January 29, 2019

  • Decided the project topic
  • Decided the meeting day and time for the project to be Tuesday 3.15PM
TODO : Develop understanding of neural networks