List your company
ResearchFormerly WizWand

Pixel Forecasting on Frozen Forecasting Unified Evaluation

22.48Mean Pixel PSNRN-WALT 500M
13.286415.673218.0620.4468Jul 18, 2025
Updated 6mo ago

Evaluation results

MethodLinks
N-WALT 500MParams=500M, Diffusion Model Samples=10, Context Frames=42025.0722.480.1922.7828.95
4DS-hType=V/M, Variant=h, Resolution=224, Params=639M, Training Time=488,282 steps, Diffusion Model Samples=10, Context Frames=42025.0720.541.422.6931.24
WALT 500MParams=500M, Diffusion Model Samples=10, Context Frames=42025.0720.41.3722.5528.95
VideoMAEType=V/M, Variant=H, Resolution=224, Params=600M, Training Time=1600 epochs, Diffusion Model Samples=10, Context Frames=42025.0720.231.183422.1328.54
4DS-eType=V/M, Variant=e, Resolution=224, Params=4B, Training Time=488,282 steps, Diffusion Model Samples=10, Context Frames=42025.0719.891.4222.0332.26
VJEPAType=V/M, Variant=L, Resolution=224, Params=300M, Training Time=90k steps, Diffusion Model Samples=10, Context Frames=42025.0719.511.1421.2624.41
VideoMAEv2Type=V/M, Variant=g, Resolution=224, Params=1B, Training Time=1200 epochs, Diffusion Model Samples=10, Context Frames=42025.0718.591.4120.8126.44
VideoPrismType=V/M/L, Variant=v_giant, Resolution=288, Params=1.1B, Training Time=500k steps, Diffusion Model Samples=10, Context Frames=42025.0717.631.0319.3223.33
DINOv2Type=I, Variant=DINOV2_L/14, Resolution=224, Params=303M, Training Time=637.5k steps, Diffusion Model Samples=10, Context Frames=42025.0716.380.85717.7819.97
SigLIPType=I/L, Variant=PaLI3-SigLIP-G-opt/14, Resolution=224, Params=2B, Training Time=107k steps, Diffusion Model Samples=10, Context Frames=42025.0713.640.3914.2916.48