Handwritten lines segmentation

• 

  Original dataset

• 

  Generated dataset

• 

  Final predictions

Inspiration

This GitHub repo https://github.com/full-stack-deep-learning/fsdl-text-recognizer-project/blob/master/lab6_sln/readme.md#lab-6-line-detection

What it does

Identify pixels of handwritten text in an image

How I built it

I build using PyTorch locally and did the final full training in the cloud. I also included everything in the repo to be able to reproduce the same results, including the scripts I used to create the cloud instance.

Challenges I ran into

Defining a good and meaningful measure of outputs quality.

Accomplishments that I'm proud of

It seem to be working quite well, there is still some more testing I need to do.

What I learned

End to end model training and generate new datasets from existing ones.

What's next for Handwritten lines segmentation

More testing to further test how well it works in the wild and create APIs to expose the model for external usage.

Built With

• numpy
• python
• pytorch