python eszsl/eszsl_gzsl.py -data AWA2/AWA1/CUB/SUN/APY -mode train -alpha [KERNEL SPACE REGULARIZER] -gamma [ATT SPACE REGULARIZER]
For testing, set mode to test and set alpha, gamma to best combination from tables below.
The numbers below are class-averaged top-1 accuracies (see ZSLGBU paper for details).
| Dataset | ZSLGBU Results | Repository Results | Hyperparams from Val |
|---|---|---|---|
| CUB | 53.9 | 53.94 | Alpha=3, Gamma=-1 |
| AWA1 | 58.2 | 56.80 | Alpha=3, Gamma=0 |
| AWA2 | 58.6 | 54.82 | Alpha=3, Gamma=0 |
| aPY | 38.3 | 38.56 | Alpha=3, Gamma=-1 |
| SUN | 54.5 | 55.69 | Alpha=3, Gamma=2 |
| Dataset | ZSLGBU Results | Repository Results | Hyperparams from Val | ||||
|---|---|---|---|---|---|---|---|
| U | S | H | U | S | H | ||
| CUB | 12.6 | 63.8 | 21.0 | 14.70 | 56.53 | 23.34 | Alpha=3, Gamma=0 |
| AWA1 | 6.6 | 75.6 | 12.1 | 5.29 | 86.84 | 9.98 | Alpha=3, Gamma=0 |
| AWA2 | 5.9 | 77.8 | 11.0 | 4.04 | 88.84 | 7.72 | Alpha=3, Gamma=0 |
| aPY | 2.4 | 70.1 | 4.6 | 2.25 | 81.07 | 4.39 | Alpha=2, Gamma=0 |
| SUN | 11.0 | 27.9 | 15.8 | 13.75 | 28.41 | 18.53 | Alpha=3, Gamma=2 |
U -> Unseen Classes; S -> Seen Classes; H-> Harmonic Mean of the 2.
[1] Original MATLAB Code by Authors
[2] https://github.com/sbharadwajj/embarrassingly-simple-zero-shot-learning