Slow, draw!
A group project of CSCI 364 Artificial Intelligence.
Han Shao, Synthia Wang, Zeo Huang
Strokes Dataset in TFRecord Format
Download the data in TFRecord from
here and upzip it.
Images Dataset in Numpy Format
The images dataset in npy format can be download through command:
gsutil cp gs://quickdraw_dataset/full/numpy_bitmap/* DESTINATION_DIRECTORY
Then npy11000.py can be used to extract the first 11,000 drawings from npy file of each category. (To reduce the size of dataset)
Recurrent Neural Networks
To run train_model.py, please use the command
python3 train_model.py python3 train_model.py
--training_data=PATH_TO_DATA_DIRECTORY/training.tfrecord-00???-of-00010
--eval_data=PATH_TO_DATA_DIRECTORY/eval.tfrecord-00???-of-00010
--classes_file=PATH_TO_DATA_DIRECTORY/training.tfrecord.classes
--model_dir=PATH_TO_MODEL_DIRECTORY
--cell_type=cudnn_lstm (Speed up training, optional)
--batch_norm=True (Speed up training, optional)
3-layer Convolutional Neural Network
To run model.py, please use the command
python3 model.py --learning_rate=LEARNING_RATE --model_dir= PATH_TO_MODEL_DIRECTORY --steps=STEPS
The npy files should be put in the directory named npy11000.
Custom AlexNet
To run model_v2.py, please use the command
python3 model_v2.py --learning_rate=LEARNING_RATE --model_dir=PATH_TO_MODEL_DIRECTORY --steps=STEPS
The npy files should be put in the directory named npy11000.