cs386

Dependencies

pip install -r requirments.txt

preprocess

Read the dataset and divide into train and valid set.

python preprocess.py

Then you can get the train.json and val.json, which contain the train and valid dataset path.

import os
import json
with open("train.json","r")as f:
	train_dataset = json.load(f)
for dir_name, filename_list in train_dataset.items():
    # label is dir_name
	for filename in filename_list:
		file_dir = os.path.join("dataset", dir_name, filename)
		print(file_dir)
# for valid set, the code is just same as above.

train

python train.py kwargs/oracle/mobilenetv2/defaults.py

Baseline: 97.36

real-world preprocessing

Use the function binarize(GrayImage, c=5) to transform a real-world gray image into binarized image, input is gray image and the output is binarized image. Parameter c is from adaptive threshold binarization, c could be larger to reduce noise but result in destroy the shape of word.

For the preprocessing process, I first apply median blur on original gray image; second, apply adaptive threshold binarization; third, apply opening (morphology transformation) to reduce snow noise and closing to link some strokes.