Google Quick, Draw! doodle recognition model implementation in Chainer
Open source for 11th Software Innovation Contest.
-
download.sh
,filelist.txt
: fetch simplifed datasets from Google Cloud Platform -
dataset.py
: Preprocess the datasets (selection and division) -
model.py
: Model definition, supports CPU and GPU (switch bygpu_id
) -
train.py
: Main functions for data reading and training -
PadIterator.py
: A special version ofSerialIterator
for padding the batch and pass its length into model -
/demo
: see/demo/README.md
Execute download.sh
to get dataset (you may need to modify it to store them in a proper folder), then dataset.py
to preprocess, train.py
to start training.
Logs, visualization and snapshots will automatically save in /result
.