notAHotDog
Remember Hot-dog or Not Hot-dog from Silicon Valley. Well this is a Not Safe For Work classifier that finds and replaces all NSFW content with cat pics.
To scrapte data, run scrape_data.sh
If you are using Windows, remove carriage return character first, before running scraping script. It will be painful otherwise
dos2unix file_name.txt
or
tr -d '\r' < filewithcarriagereturns > filewithoutcarriagereturns
To install all required packages, run
pip install -r requirements.txt
Main file is nsfw.py
We utilize InceptionV3 model trained on Imagenet as our base model for classification. Parameters for training and testing neural architecture.
'--nsfw_path', nargs='*', default=['data/amateur/IMAGES'], help='Paths to the nsfw images'
'--neutral_data_path', type=str, default='data/neutral/IMAGES', help="Path to neutral images"
'--model', type=str, default='inceptionv3', help='Model to run'
'--epochs', type=int, default=5, help="training epochs"
'--lr', type=float, default=0.001, help="learning rate for optimizer"
'--batch_size', type=int, default=32, help='Batch size for training'
'--test_batch_size', type=int, default=8, help='Batch size for testing'