Image Reconstruction on FFHQ (test)
41.8PSNREigenGS
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| EigenGSPre-training=FFHQ, ITER=100002025.03 | 41.8 | 0.99 | 99 | — | |
| EigenGSIterations=10000, Number of components=300, Gaussian points=20,0002025.03 | 41.8 | 0.99 | 99 | 12.9 | |
| EigenGSIterations=10000, Number of components=500, Gaussian points=20,0002025.03 | 41.8 | 0.99 | 99 | 12.8 | |
| EigenGSPre-training=ImageNet, ITER=100002025.03 | 41.6 | 0.99 | 99 | — | |
| EigenGSIterations=5000, Number of components=500, Gaussian points=20,0002025.03 | 40.8 | 0.99 | 99 | 6.54 | |
| EigenGSPre-training=FFHQ, ITER=50002025.03 | 40.7 | 0.99 | 98 | — | |
| EigenGSIterations=5000, Number of components=300, Gaussian points=20,0002025.03 | 40.7 | 0.99 | 98 | 5.92 | |
| EigenGSPre-training=ImageNet, ITER=50002025.03 | 40.5 | 0.99 | 97 | — | |
| GaussianImageITER=100002025.03 | 40.1 | 0.99 | 99 | — | |
| GaussianImageIterations=10000, Gaussian points=20,0002025.03 | 40.1 | 0.99 | 99 | 12.6 | |
| GaussianImageITER=50002025.03 | 39.2 | 0.99 | 98 | — | |
| GaussianImageIterations=5000, Gaussian points=20,0002025.03 | 39.2 | 0.99 | 98 | 6.36 | |
| EigenGSIterations=1000, Number of components=500, Gaussian points=20,0002025.03 | 37.7 | 0.99 | 84 | 1.14 | |
| EigenGSPre-training=FFHQ, ITER=10002025.03 | 37.5 | 0.99 | 83 | — | |
| EigenGSIterations=1000, Number of components=300, Gaussian points=20,0002025.03 | 37.5 | 0.99 | 83 | 1.31 | |
| EigenGSPre-training=ImageNet, ITER=10002025.03 | 37.2 | 0.98 | 79 | — | |
| Ours (Instance Pattern Composers)resolution=178x1782022.11 | 37.18 | — | — | — | |
| EigenGSIterations=500, Number of components=500, Gaussian points=20,0002025.03 | 36.7 | 0.98 | 78 | 0.57 | |
| EigenGSPre-training=FFHQ, ITER=5002025.03 | 36.4 | 0.98 | 76 | — | |
| EigenGSIterations=500, Number of components=300, Gaussian points=20,0002025.03 | 36.4 | 0.98 | 76 | 0.53 | |
| EigenGSPre-training=ImageNet, ITER=5002025.03 | 36.1 | 0.97 | 71 | — | |
| EigenGSIterations=100, Number of components=500, Gaussian points=20,0002025.03 | 34.8 | 0.96 | 50 | 0.25 | |
| EigenGSPre-training=FFHQ, ITER=1002025.03 | 34.4 | 0.95 | 41 | — | |
| EigenGSIterations=100, Number of components=300, Gaussian points=20,0002025.03 | 34.4 | 0.95 | 41 | 0.11 | |
| EigenGSPre-training=ImageNet, ITER=1002025.03 | 34.1 | 0.95 | 39 | — | |
| TransINRresolution=178x1782022.11 | 33.66 | — | — | — | |
| GaussianImageITER=10002025.03 | 29.4 | 0.94 | 0 | — | |
| GaussianImageIterations=1000, Gaussian points=20,0002025.03 | 29.4 | 0.94 | 0 | 0.88 | |
| EigenGSIterations=0, Number of components=500, Gaussian points=20,0002025.03 | 28.9 | 0.88 | — | — | |
| EigenGSPre-training=FFHQ, ITER=02025.03 | 28 | 0.87 | — | — | |
| EigenGSIterations=0, Number of components=300, Gaussian points=20,0002025.03 | 28 | 0.87 | — | — | |
| EigenGSPre-training=ImageNet, ITER=02025.03 | 27.2 | 0.84 | — | — | |
| GaussianImageITER=5002025.03 | 21.8 | 0.77 | 0 | — | |
| GaussianImageIterations=500, Gaussian points=20,0002025.03 | 21.8 | 0.77 | 0 | 0.45 | |
| GaussianImageITER=1002025.03 | 10.4 | 0.41 | 0 | — | |
| GaussianImageIterations=100, Gaussian points=20,0002025.03 | 10.4 | 0.41 | 0 | 0.09 | |
| AvatarTex2026.05 | — | — | — | 90 | |
| Free-UV2026.05 | — | — | — | 20 | |
| OMGTex2026.05 | — | — | — | 7 | |
| Soap2026.05 | — | — | — | 360 |