Jinhao Dong, Tong Lin
NIPS 2019
Method | Label Size: 100 | Label Size: 600 | Label Size: 1000 | Label Size: 3000 |
---|---|---|---|---|
Our re-implementation | 4.19 ± 0.22 | 2.90 ± 0.32 | 2.92 ± 0.36 | 2.33 ± 0.13 |
MarginGAN | 3.53 ± 0.57 | 3.03 ± 0.60 | 2.87 ± 0.71 | 2.06 ± 0.20 |
Table 1: Mean and standard error rates of the classifier over 5 runs. Reproduction of Table 1 from MarginGAN.
Fig. 1: Reproduction of Figure 3-a from MarginGAN.
This folder provides a re-implementation of this paper in PyTorch, developed as part of the course METU CENG 796 - Deep Generative Models. The re-implementation is provided by:
-
Can Ufuk Ertenli, ufuk.ertenli@metu.edu.tr
-
Zeynep Sonat Baltacı, sonat.baltaci@metu.edu.tr
Please see the jupyter notebook file main.ipynb for a summary of paper, the implementation notes and our experimental results.
Experiment are conducted with,
- Python 3.7.6
- PyTorch 1.5.0
- torchvision 0.6.0a0+82fd1c8
- Matplotlib 3.1.3
An additional environment.yml file is also provided.
Quantitative and qualitative results (Table 1 and Fig. 1) are obtained from the re-implementation of the preliminary study, conducted on MNIST dataset.