ocr-cnn-vlm

Datasets

The dataset have been augmented and made publicly available to allow for easy reproducibility of this work. Datasets are available on HuggingFace:

How To's

Create augmented dataset

The Street View Text Dataset

Run:

python data/augment.py

This will create data/svt1_augmented/img directory containing light, medium and heavy weather conditions (rain, fog and combined), with short description appended to the filename, e.g.:

• 19_13_combined_3x3.jpg - a 3x3 grid of all augmented images
• 19_13_combined.jpg - 3x1 grid of medium conditions
• 19_13_fog_heavy.jpg - Heavy fog
• 19_13_fog_light.jpg - Light fog
• 19_13_fog_medium.jpg - Medium fog
• 19_13_original.jpg - original image, unmodified
• 19_13_rain_fog_heavy.jpg - Heavy rain & heafy fog
• 19_13_rain_fog_light.jpg - Light rain & light fog
• 19_13_rain_fog_medium.jpg - Medium rain & medium fog
• 19_13_rain_heavy.jpg - Heavy rain
• 19_13_rain_light.jpg - Light rain
• 19_13_rain_medium.jpg - Medium rain

The script will only augment files listed in the test.xml file (249 files).