Text-to-Image Generation on MJHQ 512x512
6.1FIDPIXART-α with DC-AE-f32
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PIXART-α with DC-AE-f32Diffusion Model=PIXART-α [6], Autoencoder=DC-AE-f32, Patch Size=1, NFE=20, Throughput (image/s) Training=173, Throughput (image/s) Inference=31.27, Latency (ms)=209, Memory (GB)=23.772024.10 | 6.1 | 26.41 | |
| PIXART-α with SD-VAE-f8Diffusion Model=PIXART-α [6], Autoencoder=SD-VAE-f8 [40], Patch Size=2, NFE=20, Throughput (image/s) Training=43, Throughput (image/s) Inference=7.81, Latency (ms)=742, Memory (GB)=60.452024.10 | 6.3 | 26.36 |