Skip to content

Latest commit

 

History

History
139 lines (94 loc) · 4.89 KB

TABLE.md

File metadata and controls

139 lines (94 loc) · 4.89 KB

Comprehensive results table

Results are sorted by their absolute performance under corruption.

ResNet-50 Backbone Track

In this section only models with a ResNet 50 Backbone with Feature Pyramid Networks are listed.

Coco

Object detection:

Model Backbone box AP clean box AP corr. box %
Cascade Mask R-CNN R-50-FPN 41.2 20.7 50.2
Faster R-CNN Combined R-50-FPN 34.6 20.4 58.9
Cascade R-CNN R-50-FPN 40.4 20.1 49.7
Mask R-CNN R-50-FPN 37.3 18.7 50.1
Faster R-CNN R-50-FPN 36.3 18.2 50.2
RetinaNet R-50-FPN 35.6 17.8 50.1
Faster R-CNN Stylized R-50-FPN 21.5 14.1 65.6

Instance Segmentation:

Model Backbone mask AP clean mask AP corr. mask %
Mask R-CNN Combined R-50-FPN 32.9 19.0 57.7
Cascade Mask R-CNN R-50-FPN 35.7 17.6 49.3
Mask R-CNN R-50-FPN 34.2 16.8 49.1
Mask R-CNN Stylizes R-50-FPN 30.5 13.2 64.1

Pascal VOC

Object detection:

Model Backbone box AP50 clean box AP50 corr. box %
Faster R-CNN Combined R-50-FPN 80.4 56.2 69.9
Faster R-CNN Stylized R-50-FPN 68.0 50.0 73.5
Faster R-CNN R-50-FPN 80.5 48.6 60.4

Cityscapes

Object detection:

Model Backbone box AP clean box AP corr. box %
Faster R-CNN Combined R-50-FPN 36.3 17.2 47.4
Faster R-CNN Stylized R-50-FPN 28.5 14.7 51.5
Faster R-CNN R-50-FPN 36.4 12.2 33.4
Mask R-CNN R-50-FPN 37.5 11.7 31.1

Instance Segmentation:

Model Backbone mask AP clean mask AP corr. mask %
Mask R-CNN Combined R-50-FPN 32.1 14.9 46.3
Mask R-CNN Stylizes R-50-FPN 23.0 11.3 49.2
Mask R-CNN R-50-FPN 32.7 10.0 30.5

Unrestricted Track

Any mdel independent of it's backbone can participate in this track.

Coco

Object detection:

Model Backbone box AP clean box AP corr. box %
Hybrid Task Cascade X-101-64x4d-FPN-DCN 50.6 32.7 64.7
Faster R-CNN X-101-32x4d-FPN-DCN 43.4 26.7 61.6
Faster R-CNN X-101-64x4d-FPN 41.3 23.4 56.6
Mask R-CNN R-50-FPN-DCN 41.1 23.3 56.7
Faster R-CNN R-50-FPN-DCN 40.0 22.4 56.1
Faster R-CNN X-101-32x4d-FPN 40.1 22.3 55.5
Faster R-CNN R-101-FPN 38.5 20.9 54.2
Cascade Mask R-CNN R-50-FPN 41.2 20.7 50.2
Faster R-CNN Combined R-50-FPN 34.6 20.4 58.9
Cascade R-CNN R-50-FPN 40.4 20.1 49.7
Mask R-CNN R-50-FPN 37.3 18.7 50.1
Faster R-CNN R-50-FPN 36.3 18.2 50.2
RetinaNet R-50-FPN 35.6 17.8 50.1
Faster R-CNN Stylized R-50-FPN 21.5 14.1 65.6

Instance Segmentation:

Model Backbone mask AP clean mask AP corr. mask %
Hybrid Task Cascade X-101-64x4d-FPN-DCN 43.8 28.1 64.0
Mask R-CNN R-50-FPN-DCN 37.2 20.7 55.7
Mask R-CNN Combined R-50-FPN 32.9 19.0 57.7
Cascade Mask R-CNN R-50-FPN 35.7 17.6 49.3
Mask R-CNN R-50-FPN 34.2 16.8 49.1
Mask R-CNN Stylizes R-50-FPN 30.5 13.2 64.1

Pascal VOC

Object detection:

Model Backbone box AP50 clean box AP50 corr. box %
Faster R-CNN Combined R-50-FPN 80.4 56.2 69.9
Faster R-CNN Stylized R-50-FPN 68.0 50.0 73.5
Faster R-CNN R-50-FPN 80.5 48.6 60.4

Cityscapes

Object detection:

Model Backbone box AP clean box AP corr. box %
Faster R-CNN Combined R-50-FPN 36.3 17.2 47.4
Faster R-CNN Stylized R-50-FPN 28.5 14.7 51.5
Faster R-CNN R-50-FPN 36.4 12.2 33.4
Mask R-CNN R-50-FPN 37.5 11.7 31.1

Instance Segmentation:

Model Backbone mask AP clean mask AP corr. mask %
Mask R-CNN Combined R-50-FPN 32.1 14.9 46.3
Mask R-CNN Stylizes R-50-FPN 23.0 11.3 49.2
Mask R-CNN R-50-FPN 32.7 10.0 30.5

Methods