zilongqiu / data-centric-competition

Data-Centric AI Competition

Data-Centric AI Competition

Contest

Data-Centric AI Competition Submission Guide

Submission

zilongqiu-ai-competition CodaLab

Submit a zip file containing all the images following this structure below:

sample_submission/
    train/
        i/
        ii/
        iii/
        iv/
        ...
    val/
        i/
        ii/
        iii/
        iv/
        ...

Run training

1. Verify the train.py script folders path (directory, user_data, valid_data & test_data)
2. (optional) if using testing folder, verify the images
3. cd scripts
4. python3 train.py

Generate images

1. cd scripts/generator
2. (optional) Verify the dictionary data in the dicts/en.txt file
3. (optional) Verify the fonts in the fonts/en directory
4. make up or make rebuild
5. Makefile available commands:
   • Generate images with a specific font with make font=MY_FONT generate_w_font
   • Generate massive combinaison of all fonts make generate_all
   • Generate image for each font in specific folder make generate_grouped
   • Delete all the generated images make generate_delete
   • Copy all the generated images to testing folder make generate_to_testing
6. (optional) If using make generate_all, you need to split the testing/train images to the testing/val folder (train 80% / val 20%)

Predict image label

1. cd scripts
2. python3 predict.py ./../label_book/ii/a37502ee-ce5d-11eb-b317-38f9d35ea60f.png

Test model against the testing label_book

1. cd scripts
2. python3 predict_label_book.py ../results/baseline/model/model.h5