List your company
ResearchFormerly WizWand

Object Detection on COCO

58.7AP (Box)MViTv2-L (SoftNMS, MS)
44.76448.3825255.618Mar 19, 2021Jan 22, 2022Nov 28, 2022Oct 4, 2023Aug 8, 2024Jun 14, 2025Apr 20, 2026
Updated 5mo ago

Evaluation results

MethodLinks
MViTv2-L (SoftNMS, MS)Backbone=MViTv2-L, Pre-training=IN-21K, Training schedule=50 epochs, large-scale jittering, Inference strategy=SoftNMS and multiscale testing, Params=270 M2021.1258.7—76.764.3
MViTv2-H (IN-21K, LSJ)Backbone=MViTv2-H, Pre-training=IN-21K, Training schedule=50 epochs, large-scale jittering, FLOPs=3084 G, Params=718 M2021.1256.1—74.661
MViTv2-L (IN-21K, LSJ)Backbone=MViTv2-L, Pre-training=IN-21K, Training schedule=50 epochs, large-scale jittering, FLOPs=1519 G, Params=270 M2021.1255.8—74.360.9
EVT-LFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=157, FLOPs (G)=10292026.0455.8—74.360.4
EVT-BFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=114, FLOPs (G)=8492026.0455.5—7460.2
MViTv2-B (IN-21K)Backbone=MViTv2-B, Pre-training=IN-21K, FLOPs=814 G, Params=103 M2021.1254.9—73.859.8
CAEBackbone=ViT-L, #Epochs=1600, Pre-training Supervision=Self-supervised, Framework=Mask R-CNN, Schedule=1x2022.0254.6—75.259.9
CAE*Backbone=ViT-L, #Epochs=1600, Pre-training Supervision=Self-supervised, Framework=Mask R-CNN, Schedule=1x2022.0254.5—75.260.1
RMT-BFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=111, FLOPs (G)=8522026.0454.5—72.859
SimMIMBackbone=SwinV2-H2021.1154.4———
MViTv2-LBackbone=MViTv2-L, FLOPs=1519 G, Params=270 M2021.1254.3—73.159.1
MViTv2-BBackbone=MViTv2-B, FLOPs=814 G, Params=103 M2021.1254.1—72.958.5
MAEBackbone=ViT-L, #Epochs=1600, Pre-training Supervision=Self-supervised, Framework=Mask R-CNN, Schedule=1x2022.0254—74.359.5
OverLoCK-BFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=154, FLOPs (G)=10082026.0453.9———
CSWin-BFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=135, FLOPs (G)=10042026.0453.9—72.658.5
SimMIMBackbone=Swin-L2021.1153.8———
EVT-SFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=83, FLOPs (G)=7412026.0453.8—72.458.4
UniFormer-BFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=107, FLOPs (G)=8782026.0453.8—72.858.5
EVT-LFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=119, FLOPs (G)=5502026.0453.6—74.158.7
OverLoCK-SFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=114, FLOPs (G)=8572026.0453.6———
EVT-BFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=76, FLOPs (G)=3712026.0453.3—73.958.4
MViTv2-SBackbone=MViTv2-S, FLOPs=748 G, Params=87 M2021.1253.2—72.458
RMT-SFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=83, FLOPs (G)=7412026.0453.2—7257.8
UniRepLKNet-SFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=113, FLOPs (G)=8352026.0453———
GC ViT-BFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=146, FLOPs (G)=10182026.0452.9—71.757.8
DAT-SFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=107, FLOPs (G)=8572026.0452.7—71.757.2
CSWin-TFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=80, FLOPs (G)=7572026.0452.5—71.557.1
SimMIMBackbone=Swin-B2021.1152.3———
MViTv2-TBackbone=MViTv2-T, FLOPs=701 G, Params=76 M2021.1252.2—71.156.6
RMT-BFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=73, FLOPs (G)=3732026.0452.2—72.957
UniFormer-SFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=79, FLOPs (G)=7472026.0452.1—71.156.6
Full-precisionNetwork=Swin-S + Cascade Mask R-CNN, #Bits=32/32, Size=427.82024.1151.9———
Full-precisionNetwork=Swin-B + Cascade Mask R-CNN, #Bits=32/32, Size=579.92024.1151.9———
Swin-BBackbone=Swin-B, FLOPs=982 G, Params=145 M2021.1251.9—70.956.5
SMT-SFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=78, FLOPs (G)=7442026.0451.9—70.556.3
Swin-SFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=107, FLOPs (G)=8382026.0451.9—70.756.3
NAT-SFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=108, FLOPs (G)=8092026.0451.9—70.456.2
Swin-BFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=145, FLOPs (G)=9822026.0451.9—70.556.4
RepQ-ViT + QwTNetwork=Swin-B + Cascade Mask R-CNN, #Bits=6/6, Size=126.12024.1151.8———
Swin-SBackbone=Swin-S, FLOPs=838 G, Params=107 M2021.1251.8—70.456.3
UniRepLKNet-TFramework=Cascade Mask R-CNN, Schedule=3x+MS, Params (M)=89, FLOPs (G)=7492026.0451.8———
RepQ-ViT + QwTNetwork=Swin-S + Cascade Mask R-CNN, #Bits=6/6, Size=91.32024.1151.7———
EVT-SFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=45, FLOPs (G)=2622026.0451.6—72.456.6
RepQ-ViTNetwork=Swin-B + Cascade Mask R-CNN, #Bits=6/6, Size=112.12024.1151.5———
RepQ-ViTNetwork=Swin-S + Cascade Mask R-CNN, #Bits=6/6, Size=83.42024.1151.4———
iBOTPretrain Epochs=1600, Backbone=ViT-B/162024.1051.2———
SupervisedBackbone=Swin-L2021.1150.9———
C-JEPAPretrain Epochs=600, Backbone=ViT-B/162024.1050.7———
RMT-SFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=46, FLOPs (G)=2622026.0450.7—71.955.6
Swin-TBackbone=Swin-T, FLOPs=745 G, Params=86 M2021.1250.5—69.354.9
MILA-SFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=63, FLOPs (G)=3192026.0450.5—71.855.2
MAEPretrain Epochs=1600, Backbone=ViT-B/162024.1050.3———
InternImage-BFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=115, FLOPs (G)=5012026.0450.3—71.455.3
SupervisedBackbone=Swin-B2021.1150.2———
SupervisedBackbone=SwinV2-H2021.1150.2———
CAEBackbone=ViT-B, #Epochs=1600, Pre-training Supervision=Self-supervised, Framework=Mask R-CNN, Schedule=1x2022.0250.2—7154.9
DINOPretrain Epochs=1600, Backbone=ViT-B/162024.1050.1———
RepQ-ViT + QwTNetwork=Swin-B + Cascade Mask R-CNN, #Bits=4/4, Size=90.12024.1150———
CAE*Backbone=ViT-B, #Epochs=1600, Pre-training Supervision=Self-supervised, Framework=Mask R-CNN, Schedule=1x2022.0250—70.954.8
CSWin-SFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=54, FLOPs (G)=3422026.0450—71.354.7
RepQ-ViT + QwTNetwork=Swin-S + Cascade Mask R-CNN, #Bits=4/4, Size=64.82024.1149.9———
I-JEPAPretrain Epochs=600, Backbone=ViT-B/162024.1049.9———
CAE*Backbone=ViT-B, #Epochs=800, Pre-training Supervision=Self-supervised, Framework=Mask R-CNN, Schedule=1x2022.0249.8—70.754.6
BEiTPretrain Epochs=800, Backbone=ViT-B/162024.1049.8———
InternImage-SFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=69, FLOPs (G)=3402026.0449.7—71.154.5
ViT-Adapter-BFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=120, FLOPs (G)=8322026.0449.6—70.654
RepQ-ViTNetwork=Swin-S + Cascade Mask R-CNN, #Bits=4/4, Size=56.92024.1149.3———
RepQ-ViTNetwork=Swin-B + Cascade Mask R-CNN, #Bits=4/4, Size=76.12024.1149.3———
InternImage-TFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=49, FLOPs (G)=2702026.0449.1—70.454.1
SMT-SFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=40, FLOPs (G)=2652026.0449—70.153.4
CSWin-TFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=42, FLOPs (G)=2792026.0449—70.753.7
DetCon_BData=IN-1M, Params=250 M2021.0348.9———
Cascade RCNNBackbone=ViTAE-S-Stage, Lr Schd=1x, Params (M)=752021.0648.9———
VSSD-TFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=44, FLOPs (G)=2652026.0448.8—70.453.4
MILA-TFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=44, FLOPs (G)=2552026.0448.8—7153.6
Swin-BFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=107, FLOPs (G)=4962026.0448.6—7053.4
Full-precisionNetwork=Swin-S + Mask R-CNN, #Bits=32/32, Size=276.52024.1148.5———
SEERData=IG-1B, Params=693 M2021.0348.5———
Swin-SFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=69, FLOPs (G)=3592026.0448.5—70.253.5
MAEBackbone=ViT-B, #Epochs=1600, Pre-training Supervision=Self-supervised, Framework=Mask R-CNN, Schedule=1x2022.0248.4—69.453.1
CAE*Backbone=ViT-B, #Epochs=300, Pre-training Supervision=Self-supervised, Framework=Mask R-CNN, Schedule=1x2022.0248.4—69.252.9
MPViT-SFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=43, FLOPs (G)=2682026.0448.4—70.552.6
X101-64Backbone=ResNeXt-101-64x4d, FLOPs=972 G, Params=140 M2021.1248.3—66.452.3
iBOTBackbone=ViT-B, #Epochs=1600+, Pre-training Supervision=Self-supervised, Framework=Mask R-CNN, Schedule=1x2022.0248.2—69.752.8
Cascade RCNNBackbone=Swin-T, Lr Schd=1x, Params (M)=862021.0648.1———
X101-32Backbone=ResNeXt-101-32x4d, FLOPs=819 G, Params=101 M2021.1248.1—66.552.4
RepQ-ViT + QwTNetwork=Swin-S + Mask R-CNN, #Bits=6/6, Size=61.22024.1148———
ConvNeXt-SFramework=Mask R-CNN, Schedule=3x+MS, Params (M)=70, FLOPs (G)=3482026.0447.9—7052.7
RepQ-ViTNetwork=Swin-S + Mask R-CNN, #Bits=6/6, Size=53.32024.1147.6———
DetCon_BBackbone=ResNet-200, Fine-tune schedule=2x, Pre-trained on=ImageNet2021.0347.2———
DetCon_BBackbone=ResNet-200, Fine-tune schedule=1x, Pre-trained on=ImageNet2021.0347.1———
DeiTBackbone=ViT-B, #Epochs=300, Pre-training Supervision=Supervised, Framework=Mask R-CNN, Schedule=1x2022.0246.9—68.951
DINOBackbone=ViT-B, #Epochs=1600+, Pre-training Supervision=Self-supervised, Framework=Mask R-CNN, Schedule=1x2022.0246.8—68.650.9
Res50Backbone=ResNet-50, FLOPs=739 G, Params=82 M2021.1246.3—64.350.5
BYOLBackbone=ResNet-200, Fine-tune schedule=2x, Pre-trained on=ImageNet2021.0345.9———
SupervisedData=IN-1M, Params=250 M2021.0345.9———
BYOLBackbone=ResNet-200, Fine-tune schedule=1x, Pre-trained on=ImageNet2021.0345.6———
MoCo v3Backbone=ViT-B, #Epochs=600+, Pre-training Supervision=Self-supervised, Framework=Mask R-CNN, Schedule=1x2022.0245.5—67.149.4
MAEBackbone=ViT-B, #Epochs=300, Pre-training Supervision=Self-supervised, Framework=Mask R-CNN, Schedule=1x2022.0245.4—66.449.6
SimCLRBackbone=ResNet-200, Fine-tune schedule=2x, Pre-trained on=ImageNet2021.0345.3———