Video Instance Segmentation On Ovis 1

评估指标

AP50

AP75

AR1

AR10

mask AP

评测结果

各个模型在此基准测试上的表现结果

						Paper Title	Repository
DVIS-DAQ(VIT-L, Offline)	83.8	62.9	-	-	57.1	DVIS-DAQ: Improving Video Segmentation via Dynamic Anchor Queries
CAVIS(VIT-L, Offline)	82.6	63.5	21.2	61.8	57.1	Context-Aware Video Instance Segmentation
DVIS++(VIT-L,Offline)	78.9	58.5	-	-	53.4	DVIS++: Improved Decoupled Framework for Universal Video Segmentation
GLEE-Pro	-	55.5	-	-	50.4	General Object Foundation Model for Images and Videos at Scale
DVIS(Swin-L, Offline)	75.9	53.0	19.4	55.3	49.9	DVIS: Decoupled Video Instance Segmentation Framework
DVIS++(VIT-L, Online)	72.5	55.0	20.8	54.6	49.6	DVIS++: Improved Decoupled Framework for Universal Video Segmentation
UNINEXT (ViT-H, Online)	72.5	52.2	-	-	49.0	Universal Instance Perception as Object Discovery and Retrieval
DVIS(Swin-L, Online)	71.9	49.2	19.4	52.5	47.1	DVIS: Decoupled Video Instance Segmentation Framework
CTVIS (Swin-L)	71.5	47.5	-	-	46.9	CTVIS: Consistent Training for Online Video Instance Segmentation
RefineVIS (Swin-L, offline)	70.4	48.4	19.1	51.2	46	RefineVIS: Video Instance Segmentation with Temporal Attention Refinement	-
GRAtt-VIS (Swin-L)	69.1	47.8	19.2	49.4	45.7	GRAtt-VIS: Gated Residual Attention for Auto Rectifying Video Instance Segmentation
GenVIS (Swin-L)	69.2	47.8	18.9	49.0	45.4	A Generalized Framework for Video Instance Segmentation
NOVIS (Swin-L)	68.3	43.8	19.4	46.9	43.5	NOVIS: A Case for End-to-End Near-Online Video Instance Segmentation	-
TarViS (Swin-L)	67.8	44.6	18.0	50.4	43.2	TarViS: A Unified Approach for Target-based Video Segmentation
ROVIS (Swin-L)	64.7	42.6	18.4	49.1	42.6	Robust Online Video Instance Segmentation with Track Queries
MDQE(SwinL)	67.8	44.3	18.3	46.5	42.6	MDQE: Mining Discriminative Query Embeddings to Segment Occluded Instances on Challenging Videos
IDOL (Swin-L)	65.7	45.2	17.9	49.6	42.6	In Defense of Online Models for Video Instance Segmentation
UniVS(Swin-L)	-	-	-	-	41.7	UniVS: Unified and Universal Video Segmentation with Prompts as Queries
DVIS++(R50, Offline)	68.9	40.9	16.8	47.3	41.2	DVIS++: Improved Decoupled Framework for Universal Video Segmentation
BoxVIS(Swin-L & Box-sup)	68.4	39.9	-	-	40.6	BoxVIS: Video Instance Segmentation with Box Annotations

0 of 44 row(s) selected.

Command Palette

Video Instance Segmentation On Ovis 1

评估指标

评测结果