Skip to content
ModelsThatRun

Instance segmentation: General objects

Masks for everyday object classes.

12510205010020050010003035404550Latency on RTX 4090, ms, log scale (measured)mask AP, COCO val2017 (measured)large-640large-1280smlxnanosmallmediumlargexlarge2xlargensmlxnsmlxnsmlx
● Commercial OK● Conditional● Non-commercialBest trade-off so farInput sizes differ: compare latency with care

All general objects models

Each model runs at its official default input, shown under its latency. A smaller input is less work, so compare latencies across input sizes with care.

ModelAccuracy (paper under it)4090 latency (input under it)VRAM peakLicense
EoMT2 sizes: large-640, large-128044.8 mask AP45.2 COCO mAP (mask) paper44.7 ms640x6406.4 GiBCommercial OK
GTR-Seg4 sizes: s, m, l, x44.9 mask AP45 COCO mask AP paper35.6 ms640x640677 MiBCommercial OK
RF-DETR-Seg6 sizes: nano, small, medium, large, xlarge, 2xlarge40.1 mask AP40.3 COCO AP50:95 (mask) paper12.5 ms312x312887 MiBCommercial OK
YOLO11-seg5 sizes: n, s, m, l, x31.8 mask AP32 COCO mAPmask 50-95 paper14.5 ms640x448596 MiBConditional
YOLO26-seg5 sizes: n, s, m, l, x33.5 mask AP33.9 COCO mAPmask 50-95 (e2e) paper15.5 ms640x448579 MiBConditional
YOLOv8-seg5 sizes: n, s, m, l, x30.1 mask AP30.5 COCO mAPmask 50-95 paper12.8 ms640x448590 MiBConditional