Create an ML Model with OCR on Android For Lottery Ticket Info Extraction
Budget: $1,500 – $3,000 USD
I require a skilled and experienced Machine Learning engineer to develop and implement an ML model in an Android environment. The model will read and extract data from lottery tickets targeting specifically Powerball tickets and Mega Millions tickets.
To train the model:
I have a dataset of 1300 images, 1000 of which have been annotated already on RoboFlow.
There are about 300 that still need to be annotated.
The images consist of tickets from various states that have various layouts.
An image is attached to show how the images are annotated and what information is required.
The goal is to be able to scan tickets with any number of lines and get all the required information for checking the ticket.
Key Requirements:
1. Detect and extract all the required text from the ticket.
2. Speed - it must be possible to scan a ticket in less than 1 second.
3. Accuracy - on a test set of images an accuracy of 95% or higher is expected for both segment detection and OCR
4. Size - the implementation should not require more than 20mb
Final deliverables:
You will need to deliver a demo Android app (Java preferred) that that can accurately extract the required information from lottery tickets. The app should launch the camera with a graphical overlay of the segments/objects found, and then display the extracted text from those segments/objects on a separate screen in a structured format once all the information is found.
I am looking for someone with expert experience in training models AND implementing them on mobile.
Do not apply for this position unless you have the relevant experience in both.
All data MUST be kept private. You MUST advise if you need to collaborate with anyone else to complete the work.
As a developer myself I will also appreciate that you communicate with some detail as the work progresses, and let me know what tools you are using etc. as I am interested to learn.
To apply for this work, please outline:
1. Your plan on how to annotate the remaining images
2. How you will train a model that will work on Android (be specific, model type etc)
3. How you will implement it on Android
4. Provide examples of similar projects you have worked on
Please feel free to mention any unique solutions you may have to solve this problem that are not mentioned here, or ask any questions needed.
To train the model:
I have a dataset of 1300 images, 1000 of which have been annotated already on RoboFlow.
There are about 300 that still need to be annotated.
The images consist of tickets from various states that have various layouts.
An image is attached to show how the images are annotated and what information is required.
The goal is to be able to scan tickets with any number of lines and get all the required information for checking the ticket.
Key Requirements:
1. Detect and extract all the required text from the ticket.
2. Speed - it must be possible to scan a ticket in less than 1 second.
3. Accuracy - on a test set of images an accuracy of 95% or higher is expected for both segment detection and OCR
4. Size - the implementation should not require more than 20mb
Final deliverables:
You will need to deliver a demo Android app (Java preferred) that that can accurately extract the required information from lottery tickets. The app should launch the camera with a graphical overlay of the segments/objects found, and then display the extracted text from those segments/objects on a separate screen in a structured format once all the information is found.
I am looking for someone with expert experience in training models AND implementing them on mobile.
Do not apply for this position unless you have the relevant experience in both.
All data MUST be kept private. You MUST advise if you need to collaborate with anyone else to complete the work.
As a developer myself I will also appreciate that you communicate with some detail as the work progresses, and let me know what tools you are using etc. as I am interested to learn.
To apply for this work, please outline:
1. Your plan on how to annotate the remaining images
2. How you will train a model that will work on Android (be specific, model type etc)
3. How you will implement it on Android
4. Provide examples of similar projects you have worked on
Please feel free to mention any unique solutions you may have to solve this problem that are not mentioned here, or ask any questions needed.