search

the-full-stack / fsdl-text-recognizer-2022

Source of the FSDL 2022 labs, which are at https://github.com/full-stack-deep-learning/fsdl-text-recognizer-2022-labs
https://fullstackdeeplearning.com/course
MIT License
81 stars 26 forks source link

improvements to IAM dataset code #41

Closed charlesfrye closed 2 years ago

charlesfrye commented 2 years ago

incorporates #31 and #30 and #21 and adds in some reorganization code

making a PR to trigger the CI testing via GitHub Actions, which starts from a clean slate

charlesfrye commented 2 years ago

FYI, this is still a dont-merge draft until i get a chance to kick the tires on overfitting and on the IAMLines dataset

charlesfrye commented 2 years ago

a sniff test on a few hours of single GPU training look okay -- W&B Report

charlesfrye commented 2 years ago

the first four commits are a nuisance -- trying out and then reverting a cacheing change, with some bonus docs commits

the last three are more interesting -- i rationalized the scaling of images by making it primarily the responsibility of IAMLines and connecting it with IAMParagraphs

charlesfrye commented 2 years ago

@srbhchandra merge party 🦀 🦀 🦀