Semantic Segmentation on ADE20k (512x512)
54.86mIoUJetViT-DINOv3
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| JetViT-DINOv3Size(Spec)=7B (2 FA), Latency (ms)=221.08, Throughput (samples/s)=4.802026.05 | 54.86 | 83.7 | |
| DINOv3Size(Spec)=7B, Latency (ms)=316.79, Throughput (samples/s)=3.262026.05 | 54.72 | 83.55 | |
| SegViTBackbone=ViT-Large, GFLOPs=624, FPS=382025.09 | 53 | — | |
| DINOv3Size(Spec)=Large, Latency (ms)=35.07, Throughput (samples/s)=29.532026.05 | 52.57 | 82.74 | |
| JetViT-DINOv3Size(Spec)=Large (2 FA), Latency (ms)=18.77, Throughput (samples/s)=57.902026.05 | 52.31 | 82.75 | |
| SegViT + DToPBackbone=ViT-Large, GFLOPs=465, FPS=62025.09 | 52.3 | — | |
| SegViT + CTSBackbone=ViT-Large, GFLOPs=410, FPS=412025.09 | 52 | — | |
| SegViT + STEP@[18]τ-6899Backbone=ViT-Large, GFLOPs=395, FPS=222025.09 | 51.7 | — | |
| SegViT + CTS & DToPBackbone=ViT-Large, GFLOPs=334, FPS=12.52025.09 | 51.2 | — | |
| SegViT + STEP@[8,16]τ-6899Backbone=ViT-Large, GFLOPs=224, FPS=142025.09 | 51.2 | — | |
| AM-RADIO v3Size(Spec)=Large, Latency (ms)=33.24, Throughput (samples/s)=31.152026.05 | 51.2 | 82.61 | |
| SegViT + STEP@[8,16]τ-4999Backbone=ViT-Large, GFLOPs=334, FPS=152025.09 | 50.8 | — | |
| SegViT + STEP@[18]τ-4999Backbone=ViT-Large, GFLOPs=261, FPS=26.52025.09 | 50.4 | — | |
| DINOv3Size(Spec)=Base, Latency (ms)=12.86, Throughput (samples/s)=81.522026.05 | 50.32 | 82.59 | |
| JetViT-DINOv3Size(Spec)=Base (2 FA), Latency (ms)=8.27, Throughput (samples/s)=133.682026.05 | 49.85 | 82.53 | |
| JetViT-DINOv3Size(Spec)=Base (1 FA), Latency (ms)=7.69, Throughput (samples/s)=145.452026.05 | 49.36 | 82.4 | |
| AM-RADIO v3Size(Spec)=Base, Latency (ms)=12.08, Throughput (samples/s)=87.592026.05 | 49.1 | 81.87 | |
| JetViT-DINOv3Size(Spec)=Base (0 FA), Latency (ms)=7.37, Throughput (samples/s)=153.942026.05 | 48.84 | 82.16 | |
| SEAOTTER-FTOperating point=n = 12, Transmit CR=256, Storage CR=46, Encode (MPx/s)=256.37, Decode (MPx/s)=67.97, Protocol=Fine-tuned2026.06 | 32.77 | — | |
| AVIFOperating point=q = 5, Transmit CR=279, Storage CR=279, Encode (MPx/s)=5.43, Decode (MPx/s)=19.792026.06 | 32.75 | — | |
| AVIF (max-speed)Operating point=q = 6, Transmit CR=238, Storage CR=238, Encode (MPx/s)=32.77, Decode (MPx/s)=19.552026.06 | 32.51 | — | |
| WaLLoCOperating point=p = 11, Transmit CR=248, Storage CR=248, Encode (MPx/s)=40.31, Decode (MPx/s)=3.242026.06 | 30.51 | — | |
| SEAOTTER-ZSOperating point=n = 12, Transmit CR=256, Storage CR=20, Encode (MPx/s)=256.37, Decode (MPx/s)=65.35, Protocol=Zero-shot2026.06 | 30.09 | — | |
| FRAPPEOperating point=n = 12, Transmit CR=256, Storage CR=256, Encode (MPx/s)=256.37, Decode (MPx/s)=0.682026.06 | 29.09 | — |