List your company
ResearchFormerly WizWand

Semantic Segmentation on ADE20K 2016 (val)

59.5mIoUEoMT
55.96456.88257.858.718Mar 26, 2026
Updated 7mo ago

Evaluation results

MethodLinks
EoMTBackbone=ViT-L, Pre-training=DINOv3, Encoder=Finetuned, Input size=512², Params=315M, GFLOPs=721, FPS=1512026.0359.5
Mask2FormerBackbone=ViT-Adapter-L, Pre-training=DINOv3, Encoder=Frozen, Input size=512², Params=340M, GFLOPs=879, FPS=402026.0358.7
PMTBackbone=ViT-L, Pre-training=DINOv3, Encoder=Frozen, Input size=512², Params=357M, GFLOPs=823, FPS=1282026.0358.5
EoMTBackbone=ViT-L, Pre-training=DINOv2, Encoder=Finetuned, Input size=512², Params=316M, GFLOPs=721, FPS=1542026.0358.4
OneFormer†Backbone=DiNAT-L, Pre-training=IN21K, Encoder=Finetuned, Input size=896², Params=223M, GFLOPs=678, FPS=192026.0358.1
Mask2FormerBackbone=ViT-Adapter-L, Pre-training=DINOv2, Encoder=Frozen, Input size=512², Params=341M, GFLOPs=879, FPS=412026.0357.5
OneFormer†Backbone=ConvNext-XL, Pre-training=IN21K, Encoder=Finetuned, Input size=640², Params=373M, GFLOPs=607, FPS=212026.0357.4
PMTBackbone=ViT-L, Pre-training=DINOv2, Encoder=Frozen, Input size=512², Params=358M, GFLOPs=823, FPS=1352026.0357.2
MaskDINO†Backbone=Swin-L, Pre-training=IN21K, Encoder=Finetuned, Input size=640², Params=223M2026.0356.6
Mask2Former†Backbone=Swin-L, Pre-training=IN21K, Encoder=Finetuned, Input size=640², Params=216M, FPS=332026.0356.1