Image Generation on FFHQ 128x128 (test)
24.37FIDLLSA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LLSABackbone=DiT-S, Training Epochs=20, Top-K value=82025.12 | 24.37 | 436.4 | |
| Full AttentionBackbone=DiT-S, Training Epochs=202025.12 | 24.91 | 188.88 | |
| SLABackbone=DiT-S, Training Epochs=20, Top-K value=202025.12 | 25.73 | 365.48 | |
| VSABackbone=DiT-S, Training Epochs=20, Top-K value=202025.12 | 26.91 | 421.02 |