Joaquim de Curtò, Irene de Zarzà, Hong Yan, Carlos T. Calafate
In this paper, we bring forward the use of the recently developed Signature Transform as a way to measure the similarity between image distributions and provide detailed acquaintance and extensive evaluations. We are the first to pioneer RMSE and MAE Signature, along with log-signature as an alternative to measure GAN convergence, a problem that has been extensively studied. We are also forerunners to introduce analytical measures based on statistics to study the goodness of fit of the GAN sample distribution that are both efficient and effective. Current GAN measures involve lots of computation normally done at the GPU and are very time consuming. In contrast, we diminish the computation time to the order of seconds and computation is done at the CPU achieving the same level of goodness. Lastly, a PCA adaptive t-SNE approach, which is novel in this context, is also proposed for data visualization.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Image Generation | AFHQ Wild | MAE Signature | 25578 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | AFHQ Wild | MAE log-signature | 20359 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | AFHQ Wild | RMSE Signature | 33306 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | AFHQ Wild | RMSE log-signature | 26622 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | AFHQ Dog | MAE Signature | 30441 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | AFHQ Dog | MAE log-signature | 24612 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | AFHQ Dog | RMSE Signature | 38861 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | AFHQ Dog | RMSE log-signature | 31686 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | AFHQ Cat | MAE Signature | 45968 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | AFHQ Cat | MAE log-signature | 22297 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | AFHQ Cat | RMSE Signature | 61450 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | AFHQ Cat | RMSE log-signature | 29201 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | MetFaces | MAE Signature | 19872 | t-Stylegan3-ada (NVIDIA pre-trained) |
| Image Generation | MetFaces | MAE log-signature | 13761 | t-Stylegan3-ada (NVIDIA pre-trained) |
| Image Generation | MetFaces | RMSE Signature | 30894 | t-Stylegan3-ada (NVIDIA pre-trained) |
| Image Generation | MetFaces | RMSE log-signature | 21560 | t-Stylegan3-ada (NVIDIA pre-trained) |
| Image Generation | MetFaces | MAE Signature | 22799 | r-Stylegan3-ada (NVIDIA pre-trained) |
| Image Generation | MetFaces | MAE log-signature | 16539 | r-Stylegan3-ada (NVIDIA pre-trained) |
| Image Generation | MetFaces | RMSE Signature | 34977 | r-Stylegan3-ada (NVIDIA pre-trained) |
| Image Generation | MetFaces | RMSE log-signature | 24707 | r-Stylegan3-ada (NVIDIA pre-trained) |
| Image Generation | MetFaces | MAE Signature | 23428 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | MetFaces | MAE log-signature | 18071 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | MetFaces | RMSE Signature | 33247 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | MetFaces | RMSE log-signature | 25685 | Stylegan2-ada (NVIDIA pre-trained) |
| Image Generation | NASA Perseverance | MAE Signature | 9086 | Stylegan2-ada |
| Image Generation | NASA Perseverance | MAE log-signature | 5717 | Stylegan2-ada |
| Image Generation | NASA Perseverance | RMSE Signature | 11601 | Stylegan2-ada |
| Image Generation | NASA Perseverance | RMSE log-signature | 7397 | Stylegan2-ada |