> ML_BENCHMARK // COCO-INSTANCE-SEGMENTATION-MAP_v1.0
MS COCO Instance Segmentation (Mask AP)
Instance Segmentation · task-image-segmentation · active-headroom
Instance Segmentationactive-headroomhigh-vram-dependent
Evaluation Protocol
Pixel-level polygon mask overlap evaluated against ground truth segmentation masks across 80 object classes.
Baseline & Metrics
Canonical Baseline:Mask R-CNN (ResNet-50): 37.1 Mask AP | Mask2Former (Swin-L): 56.1 Mask AP
Evaluated Metrics:
Mask mAP (0.50:0.95)Mask AP50Mask AP75
Contamination & Leakage Risks
Strict test-dev evaluation.
Reproducibility Concerns
Mask RLE encoding resolution and threshold calibration.
CONNECTED DATASETMS COCO 2017 (Common Objects in Context)Computer Vision & Scene Understanding · 118,287 training images, 80 object categories, 5,000 validation images
View Dataset Spec