lightly-ocr

lightly's backend - receipt to text 📉

OCR can be found in /ocr, database/network controller can be found in /ingress

table of content

credits
requirements
instruction
develop
structure
todo
tldr

credits

requirements

recommends pipenv for virtualenv, refers to pypa/pipenv for installation
python >=3.7
built with pytorch

instruction

build local docker image with [sudo] docker build [--gpus=all] -f [PATH of Dockerfile] -t aar0npham/lightly-ocr:latest ocr
- for --gpus=all requires nvidia-docker
to test the images after build do [sudo] docker run -d -p 5000:5000 aar0npham/lightly-ocr:latest
refers to @66171c80 for MORAN
run bash scripts/download_model.sh to get the pretrained model (included in docker images)
to test the model run: python ocr/pipeline.py --img [IM_PATH]
to train your own model refers to here. Train loops are located in ocr/train/

develop

make sure the repository is up to date with git pull origin master
create a new branch BEFORE working with git checkout -b feats-branch_name where feats is the feature you are working on and the branch-name is the directory containing that features.

e.g: git checkout -b YOLO-ocr means you are working on YOLO inside ocr
make your changes as you wish
git status to show your changes and then following with git add . to stage your changes
commit these changes with git commit -m "describe what you do here"

if you have more than one commit you should rebase/squash small commits into one

git status to show the amount of your changes comparing to HEAD:

Your branch is ahead of 'origin/master' by n commit. where n is the number of your commit
git rebase -i HEAD~n to changes commit, REMEMBER -i
Once you enter the interactive shell pick your first commit and squash all the following commits after that
after saving and exits edit your commit message once the new windows open describe what you did
more information here

push these changes with git push origin feats-branch_name, git branch to check which branch you are on
then make a pull-request on github!

structure

overview in src as follows:

./
├── save_models/                 # location for model save
├── notebook/                    # contains playground for testing
├── modules/                     # contains core file for setting up models
├── train/                       # code to train specific model
├── test/                        # contains unit testing file
├── tools/                       # contains tools to generate dataset/image processing etc.
├── Dockerfile                   # Dockerfile for creating container
├── requirements.txt             # contains modules using for import
├── requirements-dev.txt         # contains modules that is not included in the docker container
├── config.yml                   # config.yml
├── convert.py                   # convert model to .onnx file format
├── model.py                     # contains model constructed from `modules`
├── net.py                       # contains detector and recognizer segment of OCR
├── server.py                    # run basic server with `flask`
├── pipeline.py                  # end-to-end OCR
└── torch2onnx.py                # conversion to .onnx (future use with onnx.js) _WIP_