During training, the results will be saved as an html file that include the hyper-parameters, the loss and accuracy, and a figure of loss value and accuracy value
Python > 3.6
Pytorch > 1.0.0
Cuda10.0
cudnn 7.4
Download data from here and unzip it unzip data.zip. The data used in this paper is provided by Xian(https://www.mpi-inf.mpg.de/departments/computer-vision-and-machine-learning/research/zero-shot-learning/zero-shot-learning-the-good-the-bad-and-the-ugly/)
GZSL performance evaluated under the setting proposed in Zero-Shot Learning - A Comprehensive Evaluation of the Good, the Bad and the Ugly.Yongqin Xian, Christoph H. Lampert, Bernt Schiele, Zeynep Akata.
ResNet-101 feature, GBU split, averaged per class accuracy.
| Model | AWA1 ts | AWA1 tr | AWA1 H | AWA2 ts | AWA2 tr | AWA2 H | CUB ts | CUB tr | CUB H | SUN ts | SUN tr | SUN H |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DEM | 32.8 | 84.7 | 47.3 | 30.5 | 86.4 | 45.1 | 19.6 | 54.0 | 13.6 | 20.5 | 34.3 | 25.6 |
| LESAE | 19.1 | 70.2 | 30.0 | 21.8 | 70.6 | 33.3 | 24.3 | 53.0 | 33.3 | 21.9 | 34.7 | 26.9 |
| TVN | 27.0 | 67.9 | 38.6 | – | – | – | 26.5 | 62.3 | 37.2 | 22.2 | 38.3 | 28.1 |
| ZSKL | 18.3 | 79.3 | 29.8 | 18.9 | 82.7 | 30.8 | 24.2 | 63.9 | 35.1 | 21.0 | 31.0 | 25.1 |
| CSSD | 34.7 | 87.1 | 49.6 | – | – | – | 19.1 | 62.7 | 29.3 | – | – | – |
| BZSL | 19.9 | 23.9 | 21.7 | – | – | – | 18.9 | 25.1 | 20.9 | 17.3 | 17.6 | 17.4 |
| UVDS | 15.3 | 79.5 | 25.7 | – | – | – | 23.8 | 76.5 | 36.3 | – | – | – |
| DCN | 25.5 | 84.2 | 39.1 | – | – | – | 28.4 | 60.7 | 38.7 | 25.5 | 37.0 | 30.2 |
| NIWT | – | – | – | 42.3 | 38.8 | 40.5 | 20.7 | 41.8 | 27.7 | – | – | – |
| RN | 31.4 | 91.3 | 46.7 | 30.0 | 93.4 | 45.3 | 38.1 | 61.1 | 47.0 | – | – | – |
| ours | 48.5 | 59.8 | 53.6 | 52.4 | 60.9 | 56.3 | 30.2 | 63.4 | 40.9 | 32.2 | 59.0 | 41.6 |
C. Li, X. Ye, H. Yang, Y. Han, X. Li and Y. Jia, "Generalized Zero Shot Learning via Synthesis Pseudo Features," in IEEE Access. doi: 10.1109/ACCESS.2019.2925093