Research Article
CPGAN : An Efficient Architecture Designing for Text-to-Image Generative Adversarial Networks Based on Canonical Polyadic Decomposition
Table 1
Experimental results of different rank ratios for the CPGAN in Oxford-102.
| Ratio | FID | IS | Flops | #Parameters | 0.1 | 158.74 | 3.26 ± 0.05 | | | 0.2 | 98.06 | 3.43 ± 0.06 | | | 0.3 | 85.08 | 2.98 ± 0.05 | | | 0.4 | 81.17 | 2.96 ± 0.04 | | | 0.5 | 74.69 | 3.04 ± 0.05 | | | 0.6 | 77.59 | 3.54 ± 0.06 | | | 0.7 | 76.50 | 2.80 ± 0.06 | | | 0.8 | 79.04 | 3.56 ± 0.05 | | | 0.9 | 76.97 | 3.17 ± 0.05 | | | 1.0 | 77.07 | 3.23 ± 0.05 | | |
|
|