YOLO

Training on Form Dataset

Dependencies

• Python3
• tensorflow 1.13.1
• numpy
• opencv 3
• Cython

Getting started

(Omit python keyword if not on windows)

1. Download Form Dataset images and put it into image/train
2. Download Form Dataset annotations and put it into annotation/train
3. Run python setup.py build_ext --inplace
4. Run python flow --model cfg/yolo-forms.cfg --train --annotation data/annotation/train --dataset data/image/train
   • You can add the --load [path_to_file] flag if you have pre-trained weights to use.

Notes

• annotation/test contains ALL xml files, while annotation/test_* contains subsets of those xml files to train on smaller datasets.
• Images and Annotations can be found at: https://github.com/MrPanda1/FormDataset
• Darkflow can be found at: https://github.com/thtrieu/darkflow
• Pretrained weights:
  • https://pjreddie.com/media/files/yolov3.weights
  • https://drive.google.com/drive/folders/0B1tW_VtY7onidEwyQ2FtQVplWEU