- Dependency =============
Image DB and Streamer https://github.com/aaalgo/picpac
Wrappers for various frameworks https://github.com/aaalgo/xnn
Caffe with PicPac support https://github.com/aaalgo/caffe-picpac
- Data Import ==============
Rename the 10 folders under train as 0, 1, ..., 9.
Import the data with
# scan a directory for images and create a database
# named db
# the directory train should contain sub-directories
# named 0, 1, ... (n-1), with n = # category
picpac-import -f 2 train db
- Training with Caffe ======================
Update train_val.prototxt and set the parameter path under picpac_param to the above produced db file.
The score 0.74081 was produced with quick_solver with 18634 iterations.
- Prediction =============
Copy the caffe snapshot model to model/caffe.params. Run "submit" to produce submition file.