ResearchBenchmarksImage Generation on Mapillary Vistas unconditional 2048x2048Image Generation on Mapillary Vistas unconditional 2048x2048Follow459Throughput (Training)DC-AE-f6469170.25271.5372.75Oct 14, 2024Evaluation resultsMethodMethodLinksThroughput (Training)Throughput (Inference)Latency (ms)Memory (GB)FID (w/o CFG)DC-AE-f64Diffusion Model=DiT-S,...Diffusion Model=DiT-S, Patch Size=1, NFE=2502024.1045910.916391159.55SD-VAE-f8Diffusion Model=DiT-S,...Diffusion Model=DiT-S, Patch Size=4, NFE=2502024.10841.643,56141.469.5