Source code for paper Fully automated multi-label image annotation by Convolutional Neural Network and Adaptive Thresholding at SoICT'16

NB: The code was intended for internal development, we have not refactored to make it ready-to-run in user-friendly fashion yet. Sorry for the inconvenience.

In essence, the "main" code is as in ./src/fullmodel_msrcv2_fixed.py. The other lengthy sources are just embarrassingly copy-and-paste copycats with appropriate amending from the main code (Yes, we were total novices at source code management and collaboration back then)

• cd to ./src/ before running the code
• t_train takes a while to generate, and will be stored in ./src/t/ to reuse.
• all evaluation metrics to be stored in ./src/metrics/
• snapshots of scoring module to be stored in ./snapshot_models/