List your company
ResearchFormerly WizWand

Image Segmentation on COCO

73.2mIoUClean
0.71219.53138.3557.169Nov 28, 2023Apr 27, 2024Sep 25, 2024Feb 24, 2025Jul 25, 2025Dec 23, 2025May 24, 2026
Updated 5mo ago

Evaluation results

MethodLinks
CleanPrompt=Box, SAM Model=SAM-Huge2026.0573.2
CleanPrompt=Box, SAM Model=SAM-Large2026.0573
Adaptformer + SAM-COBOTTrainable Parameters (K)=324.0, Backbone=ViT-Base2023.1172.2
LoRA + SAM-COBOTTrainable Parameters (K)=148.3, Backbone=ViT-Base2023.1172.1
LoRATrainable Parameters (K)=147.4, Backbone=ViT-Base2023.1171.8
AdaptformerTrainable Parameters (K)=322.7, Backbone=ViT-Base2023.1171.7
CleanPrompt=Box, SAM Model=SAM-Base2026.0571.4
UADPrompt=Box, SAM Model=SAM-Huge2026.0571.1
UADPrompt=Box, SAM Model=SAM-Large2026.0570.5
LightweightTrainable Parameters (K)=0, Backbone=ViT-Base2023.1170.4
CLIP-Guided SAM (CLIP-B + SAM-B)Total Params=197M†, Trainable Params=49M, Label Fraction=1/32 (3697)2026.0569.2
SAM 3Total Params=840M, Trainable Params=23M*, Label Fraction=1/512 (232)2026.0568.8
UADPrompt=Box, SAM Model=SAM-Base2026.0568
CLIP-Guided SAM (CLIP-B + SAM-B)Total Params=197M†, Trainable Params=49M, Label Fraction=1/64 (1849)2026.0567.7
CLIP-Guided SAM (CLIP-B + SAM-B)Total Params=197M†, Trainable Params=49M, Label Fraction=1/128 (925)2026.0565.8
CLIP-Guided SAM (CLIP-L + SAM-B)Total Params=543M, Trainable Params=126M, Label Fraction=1/512 (232)2026.0565
SAM 3Total Params=840M, Trainable Params=23M*, Zero-shot=true2026.0563.7
CLIP-Guided SAM (CLIP-B + SAM-B)Total Params=197M†, Trainable Params=49M, Label Fraction=1/256 (463)2026.0563.2
CLIP-Guided SAM (CLIP-B + SAM-H)Total Params=809M, Trainable Params=50M, Label Fraction=1/512 (232)2026.0561.2
CLIP-Guided SAM (CLIP-B + SAM-B)Total Params=197M†, Trainable Params=49M, Label Fraction=1/512 (232)2026.0560.5
DarkLLMPrompt=Box, SAM Model=SAM-Large2026.0555.5
DarkLLMPrompt=Box, SAM Model=SAM-Huge2026.0555.1
FreezeTrainable Parameters (K)=0, Backbone=ViT-Base2023.1153
CleanPrompt=Point, SAM Model=SAM-Huge2026.0548.5
CleanPrompt=Point, SAM Model=SAM-Large2026.0548.1
CleanPrompt=Point, SAM Model=SAM-Base2026.0544.9
UADPrompt=Point, SAM Model=SAM-Huge2026.0544.6
UADPrompt=Point, SAM Model=SAM-Large2026.0544.2
UADPrompt=Point, SAM Model=SAM-Base2026.0539.4
DarkLLMPrompt=Box, SAM Model=SAM-Base2026.0530.5
CLIP-Guided SAM (CLIP-B + SAM-B)Total Params=197M†, Trainable Params=49M, Zero-shot=true2026.0528
TAMBackbone=Qwen3-VL2026.0425
DarkLLMPrompt=Point, SAM Model=SAM-Huge2026.0524.3
AttnLRPBackbone=Qwen3-VL2026.0424
vStreamBackbone=Qwen3-VL2026.0424
DarkLLMPrompt=Point, SAM Model=SAM-Large2026.0522
InputGradBackbone=Qwen3-VL2026.0421
AttentionBackbone=Qwen3-VL2026.0419
DarkLLMPrompt=Point, SAM Model=SAM-Base2026.053.5