An Implementation of Attribute Label Embedding (ALE) method for Zero-Shot Learning.
Please see (!!!): https://github.com/cetinsamet/attribute-label-embedding
There are 32 classes; 20 training classes + 12 test classes.
On validation phase, only 20 training classes are used and they are split as 15 training classes + 5 test classes. Hyper-parameter tuning is performed only on this phase.
Run $python3 master.py --mode validation
On test phase, all 32 classes are used and they are split as 20 training classes + 12 test classes.
Run $python3 master.py --mode test
You can run prepareData.ipynb script to preprocess data located in APY/ directory and create ready-to-use data that will be placed in APYP/ directory. Or you can follow the [link] to download data.
Y. Xian, Christoph H. Lampert, B. Schiele, Z. Akata "Zero-Shot Learning - A Comprehensive Evaluation of the Good, the Bad and the Ugly" TPAMI (2018). [pdf] [data link]