Instance Segmentation on MS COCO 2014 2017 (val)
53.1AP Mask @ IoU=0.75RF-DETR-Seg-X
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| RF-DETR-Seg-XExtra Sup.=O365+SAM2, #Params.=38.1M, GFLOPs=260.0, Latency (ms)=15.422026.03 | 53.1 | 48.8 | 72.2 | 30.6 | 53.3 | 65.9 | |
| ECInsSeg-XExtra Sup.=-, #Params.=49.9M, GFLOPs=168.1, Latency (ms)=14.962026.03 | 52 | 48.4 | 72.2 | 26.3 | 52.7 | 71.1 | |
| YOLO26-Seg-XExtra Sup.=O365+SAM2, #Params.=62.8M, GFLOPs=313.5, Latency (ms)=15.132026.03 | 51.1 | 47 | 70.8 | 29.7 | 51.8 | 63.1 | |
| RF-DETR-Seg-LExtra Sup.=O365+SAM2, #Params.=36.2M, GFLOPs=151.1, Latency (ms)=9.422026.03 | 50.9 | 47.1 | 70.5 | 28.4 | 52.1 | 65.6 | |
| MaskDINOExtra Sup.=-, #Params.=52.1M, GFLOPs=2862026.03 | 50.7 | 46.3 | 69 | 26.1 | 49.3 | 66.1 | |
| ECInsSeg-LExtra Sup.=-, #Params.=33.6M, GFLOPs=110.8, Latency (ms)=12.562026.03 | 50.5 | 47.1 | 70.9 | 24.8 | 51.1 | 69.6 | |
| GC VIT-BBackbone=GC VIT-B, Framework=Cascade Mask R-CNN, Schedule=3x, Param (M)=146, FLOPs (G)=10182022.06 | 49.8 | 45.8 | 69.2 | — | — | — | |
| ConvNeXt-BBackbone=ConvNeXt-B, Framework=Cascade Mask R-CNN, Schedule=3x, Param (M)=146, FLOPs (G)=9642022.06 | 49.5 | 45.6 | 68.9 | — | — | — | |
| GC VIT-SBackbone=GC VIT-S, Framework=Cascade Mask R-CNN, Schedule=3x, Param (M)=108, FLOPs (G)=8662022.06 | 49.3 | 45.4 | 68.5 | — | — | — | |
| YOLO26-Seg-LExtra Sup.=O365+SAM2, #Params.=28.0M, GFLOPs=139.8, Latency (ms)=7.772026.03 | 49.2 | 45.5 | 68.7 | 27.1 | 50.4 | 62.8 | |
| ConvNeXt-SBackbone=ConvNeXt-S, Framework=Cascade Mask R-CNN, Schedule=3x, Param (M)=108, FLOPs (G)=8272022.06 | 49.1 | 45 | 68.4 | — | — | — | |
| Swin-BBackbone=Swin-B, Framework=Cascade Mask R-CNN, Schedule=3x, Param (M)=145, FLOPs (G)=9822022.06 | 48.9 | 45 | 68.1 | — | — | — | |
| Swin-SBackbone=Swin-S, Framework=Cascade Mask R-CNN, Schedule=3x, Param (M)=107, FLOPs (G)=8382022.06 | 48.8 | 45 | 68.2 | — | — | — | |
| RF-DETR-Seg-MExtra Sup.=O365+SAM2, #Params.=35.7M, GFLOPs=102.0, Latency (ms)=6.352026.03 | 48.8 | 45.3 | 68.4 | 25.5 | 50.4 | 65.3 | |
| GC VIT-TBackbone=GC VIT-T, Framework=Cascade Mask R-CNN, Schedule=3x, Param (M)=85, FLOPs (G)=7702022.06 | 48.3 | 44.6 | 67.8 | — | — | — | |
| ECInsSeg-MExtra Sup.=-, #Params.=20.1M, GFLOPs=64.2, Latency (ms)=9.852026.03 | 48.3 | 45.2 | 68.2 | 22.9 | 49 | 68.1 | |
| YOLO26-Seg-MExtra Sup.=O365+SAM2, #Params.=23.6M, GFLOPs=121.5, Latency (ms)=6.532026.03 | 47.7 | 44.1 | 66.8 | 25.6 | 48.9 | 60.2 | |
| Swin-TBackbone=Swin-T, Framework=Cascade Mask R-CNN, Schedule=3x, Param (M)=86, FLOPs (G)=7452022.06 | 47.3 | 43.7 | 66.6 | — | — | — | |
| ConvNeXt-TBackbone=ConvNeXt-T, Framework=Cascade Mask R-CNN, Schedule=3x, Param (M)=86, FLOPs (G)=7412022.06 | 47.3 | 43.7 | 66.5 | — | — | — | |
| YOLO11-Seg-XExtra Sup.=-, #Params.=62.1M, GFLOPs=296.4, Latency (ms)=15.542026.03 | 47.1 | 43.8 | 68.5 | 25.7 | 49.7 | 61.4 | |
| GC VIT-TBackbone=GC VIT-T, Framework=Mask R-CNN, Schedule=3x, Param (M)=48, FLOPs (G)=2912022.06 | 46.7 | 43.2 | 67 | — | — | — | |
| YOLOv8-Seg-XExtra Sup.=-, #Params.=71.8M, GFLOPs=344.1, Latency (ms)=16.202026.03 | 46.5 | 43.4 | 67.1 | 25.6 | 48.9 | 60.4 | |
| YOLO11-Seg-LExtra Sup.=-, #Params.=27.6M, GFLOPs=132.2, Latency (ms)=10.552026.03 | 46.2 | 42.9 | 67 | 24.6 | 48.7 | 60.6 | |
| ECInsSeg-SExtra Sup.=-, #Params.=10.3M, GFLOPs=33.1, Latency (ms)=6.962026.03 | 46 | 43 | 65.7 | 20.8 | 46.3 | 65.9 | |
| RF-DETR-Seg-SExtra Sup.=O365+SAM2, #Params.=33.7M, GFLOPs=70.6, Latency (ms)=4.812026.03 | 45.9 | 43.1 | 66.2 | 21.9 | 48.5 | 64.1 | |
| YOLOv8-Seg-LExtra Sup.=-, #Params.=46.0M, GFLOPs=220.5, Latency (ms)=12.292026.03 | 45.5 | 42.6 | 66.2 | 23.9 | 47.9 | 59.5 | |
| X101-32Backbone=X101-32, Framework=Cascade Mask R-CNN, Schedule=3x, Param (M)=101, FLOPs (G)=8192022.06 | 45.2 | 41.6 | 63.9 | — | — | — | |
| X101-64Backbone=X101-64, Framework=Cascade Mask R-CNN, Schedule=3x, Param (M)=140, FLOPs (G)=9722022.06 | 45.1 | 41.7 | 64 | — | — | — | |
| Swin-TBackbone=Swin-T, Framework=Mask R-CNN, Schedule=3x, Param (M)=48, FLOPs (G)=2672022.06 | 44.9 | 41.6 | 65.1 | — | — | — | |
| ConvNeXt-TBackbone=ConvNeXt-T, Framework=Mask R-CNN, Schedule=3x, Param (M)=48, FLOPs (G)=2622022.06 | 44.9 | 41.7 | 65 | — | — | — | |
| YOLO11-Seg-MExtra Sup.=-, #Params.=22.4M, GFLOPs=113.2, Latency (ms)=9.182026.03 | 44.5 | 41.5 | 65.1 | 23 | 47.2 | 59.1 | |
| DeiT-Small/16Backbone=DeiT-Small/16, Framework=Cascade Mask R-CNN, Schedule=3x, Param (M)=80, FLOPs (G)=8892022.06 | 44.3 | 41.4 | 64.2 | — | — | — | |
| YOLOv8-Seg-MExtra Sup.=-, #Params.=27.3M, GFLOPs=110.2, Latency (ms)=9.612026.03 | 43.8 | 40.8 | 63.6 | 22 | 46 | 58.2 | |
| ResNet-50Backbone=ResNet-50, Framework=Cascade Mask R-CNN, Schedule=3x, Param (M)=82, FLOPs (G)=7392022.06 | 43.4 | 40.1 | 61.7 | — | — | — | |
| YOLO26-Seg-SExtra Sup.=O365+SAM2, #Params.=10.4M, GFLOPs=34.2, Latency (ms)=3.512026.03 | 43 | 40 | 61.5 | 21 | 44.5 | 57.3 | |
| YOLO11-Seg-SExtra Sup.=-, #Params.=10.1M, GFLOPs=35.5, Latency (ms)=7.202026.03 | 40.4 | 37.8 | 59.9 | 19.8 | 42.7 | 55.2 | |
| YOLOv8-Seg-SExtra Sup.=-, #Params.=11.8M, GFLOPs=42.6, Latency (ms)=7.082026.03 | 38.9 | 36.8 | 58.1 | 17.3 | 41.3 | 53.7 | |
| Matrix-SSLBackbone=ResNet-50, Protocol=Fine-tuning, Pre-training=ImageNet2023.05 | 38 | 35.6 | 57.5 | — | — | — | |
| MECBackbone=ResNet-50, Protocol=Fine-tuning, Pre-training=ImageNet2023.05 | 36.8 | 34.7 | 56.3 | — | — | — | |
| SimSiamBackbone=ResNet-50, Protocol=Fine-tuning, Pre-training=ImageNet2023.05 | 36.7 | 34.4 | 56 | — | — | — | |
| VICRegBackbone=ResNet-50, Protocol=Fine-tuning, Pre-training=ImageNet2023.05 | 36.7 | — | — | — | — | — | |
| MoCo v2Backbone=ResNet-50, Protocol=Fine-tuning, Pre-training=ImageNet2023.05 | 36.5 | 34.4 | 55.8 | — | — | — | |
| Barlow TwinsBackbone=ResNet-50, Protocol=Fine-tuning, Pre-training=ImageNet2023.05 | 36.5 | 34.3 | 56 | — | — | — | |
| SWAVBackbone=ResNet-50, Protocol=Fine-tuning, Pre-training=ImageNet2023.05 | 35.9 | 33.8 | 55.2 | — | — | — | |
| SimCLRBackbone=ResNet-50, Protocol=Fine-tuning, Pre-training=ImageNet2023.05 | 35.3 | 33.3 | 54.6 | — | — | — | |
| BYOLBackbone=ResNet-50, Protocol=Fine-tuning, Pre-training=ImageNet2023.05 | 35 | 33.2 | 54.3 | — | — | — |