List your company

Object Detection on COCO (minival)

57.6mAPFrozen-DETR (Co-DINO-4scale)*
38.46443.43248.453.368Mar 22, 2018Apr 27, 2019Jun 2, 2020Jul 8, 2021Aug 14, 2022Sep 19, 2023Oct 25, 2024

Evaluation results

MethodLinks
Frozen-DETR (Co-DINO-4scale)*Backbone=Swin-B(22k), #Epochs=12, GFLOPS=732, FPS=3.8, input size of foundation model=4482024.1057.675.863.241.461.774.3——————
Co-DINO-4scaleBackbone=Swin-B(22k), #Epochs=12, GFLOPS=513, FPS=6.22024.1056.874.962.541.760.972.8——————
Frozen-DETR (DDQ with HPR)Backbone=R50, #Epochs=12, GFLOPS=467, FPS=5.22024.1055.773.961.338.458.872.3——————
Frozen-DETR (Co-DINO-4scale)Backbone=R50, #Epochs=24, GFLOPS=400, FPS=6.52024.1054.372.959.236.65872.1——————
Frozen-DETR (Co-DINO-4scale)Backbone=R50, #Epochs=12, GFLOPS=400, FPS=6.52024.105472.459.1365871.5——————
Frozen-DETR (DINO-det-4scale)Backbone=R50, #Epochs=24, GFLOPS=400, FPS=6.52024.1053.2—5835.156.570.6——————
Frozen-DETR (MS-DETR)Backbone=R50, #Epochs=12, GFLOPS=423, FPS=6.92024.105371.557.835.155.870.8——————
DDQ with HPRBackbone=R50, #Epochs=12, GFLOPS=283, FPS=6.52024.1052.469.957.535.955.566.7——————
Co-DINO-5scaleBackbone=R50, #Epochs=12, GFLOPS=860, FPS=4.42024.1052.169.457.135.455.465.8——————
DDQ DETR 4scaleBackbone=R50, #Epochs=24, GFLOPS=249, FPS=8.62024.105269.557.235.254.965.9——————
Frozen-DETR (DINO-det-4scale)Backbone=R50, #Epochs=12, GFLOPS=400, FPS=6.52024.1051.970.456.733.854.969.3——————
AdaMixerbackbone=Swin-S, encoder=null, #epochs=36, GFLOPS=2342022.0351.371.255.734.254.667.3——————
Group DETR (DINO-det-4scale)Backbone=R50, #Epochs=36, GFLOPS=279, FPS=9.72024.1051.3——34.754.565.3——————
DINO-det-5scaleBackbone=R50, #Epochs=24, GFLOPS=860, FPS=4.42024.1051.369.15634.554.265.8——————
DAC-DETRBackbone=R50, #Epochs=24, GFLOPS=279, FPS=9.72024.1051.268.9563454.665.4——————
DINO-det-4scaleBackbone=R50, #Epochs=24, GFLOPS=279, FPS=9.72024.1050.468.354.833.353.764.8——————
H-Deformable-DETRBackbone=R50, #Epochs=36, GFLOPS=268, FPS=112024.105068.354.432.952.765.3——————
MS-DETRBackbone=R50, #Epochs=12, GFLOPS=252, FPS=10.82024.105067.354.431.653.264——————
BoTNet-200Backbone=BoT200, Epochs=72, Multi-scale jitter=[0.1, 2.0]2021.0149.771.354.6—————————
BoTNet-152Backbone=BoT152, Epochs=72, Multi-scale jitter=[0.1, 2.0]2021.0149.57154.2—————————
AdaMixerbackbone=ResNeXt-101-DCN, encoder=null, #epochs=36, GFLOPS=2142022.0349.568.953.931.352.366.3——————
DINO-det-4scaleBackbone=R50, #Epochs=12, GFLOPS=279, FPS=9.72024.104966.653.53252.363——————
AdaMixerbackbone=ResNet-101, encoder=null, #epochs=36, GFLOPS=2082022.03486752.43051.263.7——————
AdaMixerbackbone=ResNet-50, encoder=null, #epochs=36, GFLOPS=1322022.03476651.130.150.261.8——————
AdaMixerBackbone=R50, #Epochs=36, GFLOPS=132, FPS=16.62024.10476651.130.150.261.8——————
AdaMixerbackbone=ResNet-50, encoder=null, #epochs=24, GFLOPS=1322022.0346.765.950.529.749.761.5——————
GAIA 180-210BPre-train=GAIA, Scale=880, Depth=[3,4,25,4], Width=[48,48,96,256,384], FLOPs=209.8B, Latency=53ms, Scheduler=1x2021.0646.4167.9850.8829.8350.8358.32——————
Sparse R-CNNbackbone=ResNet-101, encoder=FPN, #epochs=36, GFLOPS=2502022.0346.464.649.528.348.361.6——————
ResNet101Pre-train=GAIA, Scale=800, Depth=[3,4,23,3], Width=[64, 64, 128, 256, 512], FLOPs=188.5B, Latency=51ms, Scheduler=1x2021.0646.0767.2850.6429.3150.4558.5——————
GAIA 150-180BPre-train=GAIA, Scale=800, Depth=[3,4,23,3], Width=[64,64,96,256,512], FLOPs=178.7B, Latency=47ms, Scheduler=1x2021.0646.0267.450.5228.850.3759.01——————
Conditional DETRbackbone=ResNet-101-DC5, encoder=TransformerEnc, #epochs=108, GFLOPS=2622022.0345.966.849.527.250.363.3——————
Efficient DETRbackbone=ResNet-101, encoder=DeformTransEnc, #epochs=36, GFLOPS=2892022.0345.764.149.528.249.160.2——————
GAIA 135-150BPre-train=GAIA, Scale=800, Depth=[4,6,23,3], Width=[48,48,96,192,640], FLOPs=149.1B, Latency=44ms, Scheduler=1x2021.0645.5566.8950.0828.9649.9458.01——————
GAIA 120-135BPre-train=GAIA, Scale=720, Depth=[3,4,23,3], Width=[64,64,128,192,640], FLOPs=133.9B, Latency=38ms, Scheduler=1x2021.0645.2766.6449.4727.9749.4457.89——————
Efficient DETRbackbone=ResNet-50, encoder=DeformTransEnc, #epochs=36, GFLOPS=2102022.0345.163.149.128.348.459——————
Conditional DETRbackbone=ResNet-50-DC5, encoder=TransformerEnc, #epochs=108, GFLOPS=1952022.0345.165.448.525.34962.2——————
Sparse R-CNNbackbone=ResNet-50, encoder=FPN, #epochs=36, GFLOPS=1742022.034563.448.226.947.259.5——————
Sparse R-CNNBackbone=R50, #Epochs=36, GFLOPS=174, FPS=17.82024.104563.448.226.947.259.5——————
Mask R-CNNBackbone=Conformer-S/16, Input size=(1120, 800), #Params=56.9 M, GFLOPs=457.72021.0544.9——28.748.858.6——————
DETRbackbone=ResNet-101-DC5, encoder=TransformerEnc, #epochs=500, GFLOPS=2532022.0344.964.747.723.749.562.3——————
GFL-R101 (T)Backbone=ResNet-101, Role=Teacher, Training Schedule=2x2023.0644.963.1492849.157.2——————
GAIA 105-120BPre-train=GAIA, Scale=640, Depth=[4,6,21,4], Width=[64,80,160,192,512], FLOPs=119.2B, Latency=33ms, Scheduler=1x2021.0644.4665.7147.4926.7748.3157.21——————
SMCAbackbone=ResNet-101, encoder=TransformerEnc, #epochs=50, GFLOPS=2182022.0344.465.24824.348.561——————
Frozen-DETR (DN-DETR-DC5)Backbone=R50, #Epochs=12, GFLOPS=372, FPS=8.52024.1044.464.847.723.847.764.6——————
FPNBackbone=Conformer-S/16, Input size=(1120, 800), #Params=54.2 M, GFLOPs=404.62021.0544.2——28.548.158.4——————
Anchor DETRbackbone=ResNet-50-DC5, encoder=DecoupTransEnc, #epochs=50, GFLOPS=1512022.0344.2—47.524.748.260.6——————
AdaMixerbackbone=ResNet-50, encoder=null, #epochs=12, GFLOPS=1322022.0344.163.147.829.54758.8——————
GAIA 90-105BPre-train=GAIA, Scale=560, Depth=[4,6,21,4], Width=[64,80,160,192,512], FLOPs=101.1B, Latency=30ms, Scheduler=1x2021.0644.0565.1547.9125.5548.6959.09——————
CrossKD+PKDStudent Backbone=ResNet-50, Teacher Backbone=ResNet-101, Framework=GFL, Training Schedule=1x2023.0643.96247.726.448.557——————
Deformable DETRbackbone=ResNet-50, encoder=DeformTransEnc, #epochs=50, GFLOPS=1732022.0343.862.647.726.447.158——————
Deformable DETRBackbone=R50, #Epochs=50, GFLOPS=173, FPS=13.42024.1043.862.647.726.447.158——————
SMCAbackbone=ResNet-50, encoder=TransformerEnc, #epochs=50, GFLOPS=1522022.0343.763.647.224.24760.4——————
CrossKDStudent Backbone=ResNet-50, Teacher Backbone=ResNet-101, Framework=GFL, Training Schedule=1x2023.0643.762.147.426.94856.2——————
Mask R-CNNBackbone=Conformer-S/32, Input size=(1344, 800), #Params=58.1 M, GFLOPs=341.42021.0543.6——27.546.956.5——————
GAIA 75-90BPre-train=GAIA, Scale=560, Depth=[4,6,21,4], Width=[64,80,96,192,512], FLOPs=88.1B, Latency=28ms, Scheduler=1x2021.0643.664.8147.3424.9747.8558.62——————
DETRbackbone=ResNet-50-DC5, encoder=TransformerEnc, #epochs=500, GFLOPS=1872022.0343.363.145.922.547.361.1——————
PKDStudent Backbone=ResNet-50, Teacher Backbone=ResNet-101, Framework=GFL, Training Schedule=1x2023.0643.361.346.925.247.956.2——————
DETRBackbone=R50, #Epochs=500, GFLOPS=86, FPS=27.82024.1043.363.145.922.547.361.1——————
Region Uncertainty QuantificationLabeled=coco-115, Unlabeled=coco-120, schedule=3x, strong data augmentation=true2021.1143.26247.5—————————
FPNBackbone=Conformer-S/32, Input size=(1344, 800), #Params=55.4 M, GFLOPs=288.42021.0543.1——26.846.555.8——————
LDStudent Backbone=ResNet-50, Teacher Backbone=ResNet-101, Framework=GFL, Training Schedule=1x2023.064361.646.625.54755.8——————
ResNet50Pre-train=GAIA, Scale=800, Depth=[3,4,6,3], Width=[64, 64, 128, 256, 512], FLOPs=137.4B, Latency=39ms, Scheduler=1x2021.0642.9164.4347.1228.0147.0753.81——————
GAIA 60-75BPre-train=GAIA, Scale=560, Depth=[4,6,15,4], Width=[48,80,96,192,512], FLOPs=74.4B, Latency=26ms, Scheduler=1x2021.0642.5964.0146.0724.8246.7857.42——————
Auto-FPNbackbone=ResNet-101, b&w=false2020.0342.5———————————
Humble teacherLabeled=coco-115, Unlabeled=coco-120, strong data augmentation=true2021.1142.4———————————
SKDStudent Backbone=ResNet-50, Teacher Backbone=ResNet-101, Framework=GFL, Training Schedule=1x2023.0642.360.245.924.446.755.6——————
Frozen-DETR (DAB-DETR-DC5)Backbone=R50, #Epochs=12, GFLOPS=372, FPS=8.52024.104263.244.922.445.461.1——————
Double-Head-Extbackbone=ResNet-101, b&w=false2020.0341.962.445.923.945.255.8——————
CrossKDBackbone=ResNet-50, Architecture=ATSS, Teacher Backbone=ResNet-1012023.0641.860.145.424.945.954.2——————
ResNet101Pre-train=ImageNet, Scale=800, Depth=[3,4,23,3], Width=[64, 64, 128, 256, 512], FLOPs=188.5B, Latency=51ms, Scheduler=3x2021.0641.7562.2445.7725.2845.6154.39——————
Region Uncertainty QuantificationLabeled=coco-115, Unlabeled=coco-120, schedule=3x2021.1141.76145.9—————————
DN-DETR-DC5Backbone=R50, #Epochs=12, GFLOPS=220, FPS=10.2, single-scale detector uses standard attention=true2024.1041.761.444.121.24560.2——————
ATSSBackbone=ResNet-101, Role=Teacher2023.0641.559.945.224.245.953.3——————
GIDStudent Backbone=ResNet-50, Teacher Backbone=ResNet-101, Framework=GFL, Training Schedule=1x2023.0641.559.645.224.345.753.6——————
Grid R-CNN w FPNbackbone=ResNet-1012018.1141.360.344.423.445.854.1——————
UbteacherLabeled=coco-115, Unlabeled=coco-120, strong data augmentation=true2021.1141.3———————————
CrossKDBackbone=ResNet-50, Architecture=FCOS, Teacher Backbone=ResNet-1012023.0641.360.644.225.145.552.4——————
FGDStudent Backbone=ResNet-50, Teacher Backbone=ResNet-101, Framework=GFL, Training Schedule=1x2023.0641.358.844.824.545.653——————
Main RegionStudent Backbone=ResNet-50, Teacher Backbone=ResNet-101, Framework=GFL, Training Schedule=1x2023.0641.158.744.424.144.653.6——————
Fine-GrainedStudent Backbone=ResNet-50, Teacher Backbone=ResNet-101, Framework=GFL, Training Schedule=1x2023.0641.158.844.823.345.453.1——————
R101 GNBackbone=ResNet-101, Normalization=Group Normalization (GN), Training protocol=from scratch2018.034161.144.9—————————
FCOSBackbone=ResNet-101, Role=Teacher2023.0640.8604424.244.352.4——————
DefeatStudent Backbone=ResNet-50, Teacher Backbone=ResNet-101, Framework=GFL, Training Schedule=1x2023.0640.858.644.224.344.653.7——————
FitNetsStudent Backbone=ResNet-50, Teacher Backbone=ResNet-101, Framework=GFL, Training Schedule=1x2023.0640.758.64423.744.453.2——————
Inside GT BoxStudent Backbone=ResNet-50, Teacher Backbone=ResNet-101, Framework=GFL, Training Schedule=1x2023.0640.758.644.223.144.553.5——————
IoU-Netbackbone=ResNet-101, b&w=false2020.0340.659——————————
Region Uncertainty QuantificationLabeled=coco-115, Unlabeled=coco-120, schedule=1x2021.1140.659.744.4—————————
GAIA 45-60BPre-train=GAIA, Scale=480, Depth=[4,6,8,4], Width=[48,48,96,256,384], FLOPs=59.4B, Latency=19ms, Scheduler=1x2021.0640.4461.0743.7721.6543.6655.96——————
Instant-teachingLabeled=coco-115, Unlabeled=coco-120, strong data augmentation=true2021.1140.2———————————
GFL-R50 (S)Backbone=ResNet-50, Role=Student, Training Schedule=1x2023.0640.258.443.323.34452.2——————
Mask R-CNNBackbone=ResNet-101†, Input size=(1333, 800), #Params=63.2 M, GFLOPs=348.82021.0540——22.64452.6——————
ResNet50Pre-train=ImageNet, Scale=800, Depth=[3,4,6,3], Width=[64, 64, 128, 256, 512], FLOPs=137.4B, Latency=39ms, Scheduler=3x2021.0639.7860.4742.2323.2742.5750.04——————
CrossKDBackbone=ResNet-50, Architecture=RetinaNet, Teacher Backbone=ResNet-1012023.0639.758.942.522.443.652.8——————
Grid R-CNN w FPNbackbone=ResNet-502018.1139.658.342.422.643.851.5——————
R50 GNBackbone=ResNet-50, Normalization=Group Normalization (GN), Training protocol=from scratch2018.0339.559.843.6—————————
Faster R-CNN w FPNbackbone=ResNet-1012018.1139.561.243.122.743.750.8——————
ResNet101Pre-train=ImageNet, Scale=800, Depth=[3,4,23,3], Width=[64, 64, 128, 256, 512], FLOPs=188.5B, Latency=51ms, Scheduler=1x2021.0639.4161.6243.0424.4243.6150.53——————
FPNBackbone=ResNet-101†, Input size=(1333, 800), #Params=60.5 M, GFLOPs=295.72021.0539.4——22.443.751.1——————
ATSSBackbone=ResNet-50, Role=Student2023.0639.457.642.823.642.950.3——————
STACLabeled=coco-115, Unlabeled=coco-120, strong data augmentation=true2021.1139.2———————————