Instance Segmentation on LVIS 0.5 (val)
22.1APrDe-confound-TDE
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| De-confound-TDEFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | 22.1 | 28.4 | 43 | 30.6 | 29 | 30.3 | — | — | — | — | — | — | — | — | |
| Capsule-TDEFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | 21.1 | 28.4 | 42.1 | 30.8 | 29.7 | 29.6 | — | — | — | — | — | — | — | — | |
| Cosine-TDEFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | 20.8 | 28.1 | 42.6 | 30.2 | 28.7 | 30.3 | — | — | — | — | — | — | — | — | |
| Forest R-CNNBackbone=ResNet-101-FPN, multi-scale training=true2020.08 | 20.2 | 28.2 | 44.3 | 29.7 | 29.6 | 29.6 | 21 | 36 | 42 | — | — | — | — | — | |
| Forest R-CNNBackbone=ResNet-101-FPN2020.08 | 20.1 | 26.9 | 42.2 | 28.4 | 27.9 | 28.3 | 23.6 | 43.5 | 51.8 | — | — | — | — | — | |
| Forest R-CNNBackbone=ResNet-50-FPN, multi-scale training=true2020.08 | 19.7 | 26.7 | 42 | 28.8 | 27.5 | 28.5 | 20.3 | 34.8 | 39.7 | — | — | — | — | — | |
| BALMS2020.07 | 19.6 | — | — | — | 28.9 | 27.5 | — | 27 | — | — | — | — | — | — | |
| Forest R-CNNBackbone=ResNet-50-FPN2020.08 | 18.3 | 25.6 | 40.3 | 27.1 | 26.4 | 27.6 | 18.5 | 32.7 | 41.1 | — | — | — | — | — | |
| GroupSoftmaxPre-Train=COCO, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.00012021.03 | 18 | 26.3 | — | — | 26.9 | 28.7 | — | — | — | — | — | — | — | — | |
| SimCalPre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.00012021.03 | 16.4 | 23.4 | — | — | 22.5 | 27.2 | — | — | — | — | — | — | — | — | |
| SOLOv2backbone=Res-101-FPN2020.03 | 16.3 | 26.8 | — | — | 27.6 | 30.1 | 16.8 | 35.8 | 47 | — | — | — | — | — | |
| SOLOv2backbone=Res-101 FPN2020.03 | 16.3 | 26.8 | — | — | 27.6 | 30.1 | 16.8 | 35.8 | 47 | — | — | — | — | — | |
| SOLOv2Backbone=ResNet-101-FPN, multi-scale training=true, image resampling=true2020.08 | 16.3 | 26.8 | — | — | 27.6 | 30.1 | 16.8 | 35.8 | 47 | — | — | — | — | — | |
| Balanced Softmaxvariant=multiple binary logistic regression2020.07 | 16.2 | — | — | — | 27.3 | 28.8 | — | 26.3 | — | — | — | — | — | — | |
| DisAlignPre-Train=COCO, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Cosine2021.03 | 16.2 | 27.9 | — | — | 29.3 | 30.8 | — | — | — | — | — | — | — | — | |
| DisAlignPre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Cosine2021.03 | 14.9 | 26.3 | — | — | 27.6 | 29.2 | — | — | — | — | — | — | — | — | |
| LDAM2020.07 | 14.6 | — | — | — | 25.3 | 26.3 | — | 24.1 | — | — | — | — | — | — | |
| Equalization Loss2020.07 | 14.6 | — | — | — | 27.3 | 26.6 | — | 25.2 | — | — | — | — | — | — | |
| EQLBackbone=ResNet-101-FPN2020.08 | 14.6 | 24.8 | 38.4 | 26.8 | 26.7 | 26.4 | — | — | — | — | — | — | — | — | |
| RFSPre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.00012021.03 | 14.6 | 24.3 | — | — | 24 | 28.5 | — | — | — | — | — | — | — | — | |
| Mask-RCNNbackbone=Res-50-FPN2020.03 | 14.5 | 24.4 | — | — | 21.3 | 28.4 | — | — | — | — | — | — | — | — | |
| Mask RCNNbackbone=Res-50 FPN2020.03 | 14.5 | 24.4 | — | — | 24.3 | 28.4 | — | — | — | — | — | — | — | — | |
| LWS2020.07 | 14.4 | — | — | — | 24.4 | 26.8 | — | 23.8 | — | — | — | — | — | — | |
| Softmax2020.07 | 13.6 | — | — | — | 24 | 27.3 | — | 23.7 | — | — | — | — | — | — | |
| SOLOv2backbone=Res-50-FPN2020.03 | 13.4 | 25.5 | — | — | 26.6 | 28.9 | 15.9 | 34.6 | 44.9 | — | — | — | — | — | |
| SOLOv2backbone=Res-50 FPN2020.03 | 13.4 | 25.5 | — | — | 26.6 | 28.9 | 15.9 | 34.6 | 44.9 | — | — | — | — | — | |
| Repeat Factor SamplingBackbone=ResNet-50-FPN2020.08 | 13.4 | 23.2 | — | — | 23.2 | 27.1 | — | — | — | — | — | — | — | — | |
| SOLOv2Backbone=ResNet-50-FPN, multi-scale training=true, image resampling=true2020.08 | 13.4 | 25.5 | — | — | 26.6 | 28.9 | 15.9 | 34.6 | 44.9 | — | — | — | — | — | |
| Focal Loss2020.07 | 12.8 | — | — | — | 23.5 | 27.5 | — | 23.4 | — | — | — | — | — | — | |
| Sigmoid2020.07 | 12.7 | — | — | — | 24 | 27.3 | — | 23.6 | — | — | — | — | — | — | |
| Repeat Factor SamplingBackbone=ResNet-50, Architecture=Mask R-CNN2020.03 | 12.2 | 21.3 | 34.9 | 22 | 21.5 | 24.7 | 15.2 | 27.1 | 36.1 | — | — | — | — | — | |
| Mask-RCNN*backbone=Res-50-FPN, training schedule=3x, re-implementation=true2020.03 | 12.1 | 24.6 | — | — | 25.8 | 28.1 | 18.7 | 31.2 | 38.2 | — | — | — | — | — | |
| Mask RCNN*-3xbackbone=Res-50 FPN2020.03 | 12.1 | 24.6 | — | — | 25.8 | 28.1 | 18.7 | 31.2 | 38.2 | — | — | — | — | — | |
| BaselinePre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Cosine2021.03 | 11.5 | 23.8 | — | — | 23.7 | 28.9 | — | — | — | — | — | — | — | — | |
| Class Balanced Loss2020.07 | 11.4 | — | — | — | 23.8 | 27.3 | — | 23.3 | — | — | — | — | — | — | |
| De-confoundFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | 11.4 | 25.7 | 38.5 | 27.8 | 26.1 | 30.9 | — | — | — | — | — | — | — | — | |
| EQLBackbone=ResNet-50, Architecture=Mask R-CNN2020.03 | 11.3 | 22.8 | 36 | 24.4 | 24.7 | 25.1 | 16.3 | 29.7 | 38.2 | — | — | — | — | — | |
| EQLBackbone=ResNet-50-FPN2020.08 | 11.3 | 22.8 | 36 | 24.4 | 24.7 | 25.1 | 16.3 | 29.7 | 38.2 | — | — | — | — | — | |
| EQLPre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.00012021.03 | 11.3 | 22.8 | — | — | 24.7 | 25.1 | — | — | — | — | — | — | — | — | |
| BaselinePre-Train=COCO, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Cosine2021.03 | 11 | 25.3 | — | — | 25.5 | 30.7 | — | — | — | — | — | — | — | — | |
| EQLFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | 10.3 | 24.9 | 37.9 | 26.7 | 27.3 | 27.8 | — | — | — | — | — | — | — | — | |
| DisAlignPre-Train=COCO, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Linear2021.03 | 9.4 | 25.7 | — | — | 27.6 | 29.7 | — | — | — | — | — | — | — | — | |
| Focal LossBackbone=ResNet-50, Architecture=Mask R-CNN2020.03 | 9.3 | 21 | 34.2 | 22.1 | 21 | 25.8 | 15.6 | 27.8 | 35.4 | — | — | — | — | — | |
| CosineFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | 9.3 | 25 | 37.7 | 27 | 25.5 | 30.8 | — | — | — | — | — | — | — | — | |
| Focal LossBackbone=ResNet-50-FPN2020.08 | 9.3 | 21 | 34.2 | 22.1 | 21 | 25.8 | 15.6 | 27.8 | 35.4 | — | — | — | — | — | |
| Focal LossPre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.00012021.03 | 9.3 | 21 | — | — | 21 | 25.8 | — | — | — | — | — | — | — | — | |
| CapsuleFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | 8.5 | 25.4 | 37.8 | 27.4 | 26.4 | 31 | — | — | — | — | — | — | — | — | |
| DisAlignPre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Linear2021.03 | 8.5 | 24.3 | — | — | 26.3 | 28.1 | — | — | — | — | — | — | — | — | |
| Class-balanced LossBackbone=ResNet-50, Architecture=Mask R-CNN2020.03 | 8.2 | 20.9 | 33.8 | 22.2 | 21.2 | 25.7 | 15.6 | 28.1 | 35.3 | — | — | — | — | — | |
| Class-balanced LossBackbone=ResNet-50-FPN2020.08 | 8.2 | 20.9 | 33.8 | 22.2 | 21.2 | 25.7 | 15.6 | 28.1 | 35.3 | — | — | — | — | — | |
| Class-aware SamplingBackbone=ResNet-50, Architecture=Mask R-CNN2020.03 | 7.3 | 18.5 | 31.1 | 18.9 | 19.3 | 21.9 | 13.3 | 24.3 | 30.5 | — | — | — | — | — | |
| Class-aware SamplingBackbone=ResNet-50-FPN2020.08 | 7.3 | 18.5 | 31.1 | 18.9 | 19.3 | 21.9 | 13.3 | 24.3 | 30.5 | — | — | — | — | — | |
| Sigmoid LossBackbone=ResNet-50, Architecture=Mask R-CNN2020.03 | 7.2 | 20.1 | 32.7 | 21.2 | 19.9 | 25.4 | 15 | 27.1 | 34.6 | — | — | — | — | — | |
| Softmax LossBackbone=ResNet-50, Architecture=Mask R-CNN2020.03 | 4.5 | 20.2 | 32.6 | 21.3 | 20.8 | 25.6 | 15.5 | 27.7 | 34.4 | — | — | — | — | — | |
| BaselinePre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Linear2021.03 | 3.7 | 21.2 | — | — | 21.6 | 28.4 | — | — | — | — | — | — | — | — | |
| Focal LossFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | 3.2 | 21.1 | 32.1 | 22.6 | 21.1 | 28.3 | — | — | — | — | — | — | — | — | |
| BaselinePre-Train=COCO, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.0001, Classifier=Linear2021.03 | 2.8 | 23.9 | — | — | 23.4 | 30.5 | — | — | — | — | — | — | — | — | |
| BaselineFramework=Cascade Mask R-CNN, Backbone=R101-FPN2020.09 | 2.5 | 22.6 | 33.5 | 24.4 | 23 | 30.2 | — | — | — | — | — | — | — | — | |
| BlendMaskBackbone=R-101+dcni3, resolution=56 × 562020.01 | — | 44.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| BMask R-CNNTraining dataset=COCO, Inputs=Ground truth boxes2021.03 | — | — | — | — | — | — | — | — | — | 57.8 | 46.1 | 50.8 | 56.3 | 40.7 | |
| LSTBackbone=ResNet-50-FPN2020.08 | — | 23 | 36.7 | 24.8 | — | — | — | — | — | — | — | — | — | — | |
| LSTPre-Train=ImageNet, Backbone=ResNet-50, Framework=Mask-R-CNN-FPN, Score threshold=0.00012021.03 | — | 23 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Mask R-CNNBackbone=X101-FPN, resolution=28 × 282020.01 | — | 40.7 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Mask R-CNNTraining dataset=COCO, Inputs=Ground truth boxes2021.03 | — | — | — | — | — | — | — | — | — | 51.5 | 38.3 | 45.4 | 48.7 | 29.2 | |
| PointRendBackbone=X101-FPN, resolution=224 x 2242020.01 | — | 43.4 | — | — | — | — | — | — | — | — | — | — | — | — | |
| PointRendTraining dataset=COCO, Inputs=Ground truth boxes2021.03 | — | — | — | — | — | — | — | — | — | 56.8 | 45.9 | 49.6 | 56 | 42.2 |