Quartz Energy - Mud Logs - OCR with Google Vision API

CHALLENGE OBJECTIVES

• Replace the model used in our phrase extraction tool to use the Google Vision API.
• Create documentation describing the comparison of the current results vs the results generated by the Google Vision API.

PROJECT BACKGROUND
According to Wikipedia, “Mud logging is the creation of a detailed record (well log) of a borehole by examining the cuttings of rock brought to the surface by the circulating drilling medium (most commonly drilling mud).”  Quartz Energy has provided Topcoder with a set of mud logs and we have developed an OCR algorithm to extract phrases of interest from these mud logs.
 
The client is interested in using the Google Vision API to do the phrase extraction tasks and wants to know its performance compared to the Tesseract’s model for reading the mud log files.
 

Individual Requirements

• The current OCR algorithm is using Tesseract along with some image clean up and dilation techniques. You should execute the code and generate a report on the existing performance of the tool (using a cross-section of the training data) before consuming the Google Vision APIs.
• At a minimum, your analysis should describe overall performance, the performance by phrase type (show, stain, trace) and performance by document.  You may use the test harness tool to generate a report of the results from the run.
• Then you should make the updates to the application to implement the Google Vision API and execute the application.  Then let’s execute the same analytical task defined above so we create a comparative analysis of the current model and the Google Vision version.
• If possible the same image and text preprocessing steps should be used in both versions of the application.
• The same images and lic files must be used for both runs so that the comparison makes sense.
• It's critical that your submission includes the comparison between the results from both runs, you must generate reports on your run results using the test harness too.
• We have another set of data (that's not going to be shared with you) that we'll use to test your submission for performance and accuracy, it's going to take a big percentage in the evaluation of your submissions.

  
DEVELOPMENT ASSETS

• We will provide access to the existing code base on the forum.
• We will provide test data on the forum.

 
TECHNOLOGY STACK
The following technology stack will be used as part of this code base and this challenge.

• Java SE 8
• OpenCV 3.3
• Tesseract 4.0.0 alpha
• XGBoost 0.6
• Apache Commons Imaging
• Image IO
• Google Vision API
• Google Cloud Function (if required for model)

 
SCORECARD REVIEW

• Submissions to this challenge will be reviewed internally by the team.
• Your submission will be reviewed on these requirements:
  • Challenge Spec Requirements
    • Requirements Coverage
  • Coding Standards
    • Best Practices
    • Code Quality
  • Development Requirements
    • Performance
    • Accuracy
    • Security
    • Deployment

Final Submission Guidelines

FINAL DELIVERABLES

• Submit a single zip containing the full source code. Winner will be expected to also submit via git patch and a merge request to vision_api branch.
• A brief write up explaining how to build, configure and deploy your code.
• A document that compares the results generated by Google Vision vs Tesseract on a subset of the training data
• A submission.csv file which identifies phrases from the testing images.