CenterMask : Real-Time Anchor-Free Instance Segmentation

arXiv 2019  ·  Youngwan Lee, Jongyoul Park ·

We propose a simple yet efficient anchor-free instance segmentation, called CenterMask, that adds a novel spatial attention-guided mask (SAG-Mask) branch to anchor-free one stage object detector (FCOS) in the same vein with Mask R-CNN. Plugged into the FCOS object detector, the SAG-Mask branch predicts a segmentation mask on each box with the spatial attention map that helps to focus on informative pixels and suppress noise. We also present an improved backbone networks, VoVNetV2, with two effective strategies: (1) residual connection for alleviating the optimization problem of larger VoVNet \cite{lee2019energy} and (2) effective Squeeze-Excitation (eSE) dealing with the channel information loss problem of original SE. With SAG-Mask and VoVNetV2, we deign CenterMask and CenterMask-Lite that are targeted to large and small models, respectively. Using the same ResNet-101-FPN backbone, CenterMask achieves 38.3%, surpassing all previous state-of-the-art methods while at a much faster speed. CenterMask-Lite also outperforms the state-of-the-art by large margins at over 35fps on Titan Xp. We hope that CenterMask and VoVNetV2 can serve as a solid baseline of real-time instance segmentation and backbone network for various vision tasks, respectively. The Code is available at https://github.com/youngwanLEE/CenterMask.

PDF Abstract

Datasets


Results from the Paper


 Ranked #1 on Object Detection on COCO test-dev (Hardware Burden metric)

     Get a GitHub badge
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Object Detection COCO minival CenterMask+VoVNetV2-99 (single-scale) box AP 45.6 # 67
APS 29.2 # 16
APL 58.8 # 29
Object Detection COCO minival CenterMask+VoVNet99 (multi-scale) box AP 48.6 # 45
AP50 67.8 # 18
Object Detection COCO minival Mask R-CNN (VoVNetV2-99, single-scale) box AP 44.9 # 72
APS 28.5 # 20
APL 57.7 # 37
Instance Segmentation COCO minival CenterMask-VoVNetV2-99-3x mask AP 40.2 # 47
Object Detection COCO minival CenterMask+X101-32x8d (single-scale) box AP 44.4 # 83
APS 26.7 # 30
APL 57.1 # 41
Instance Segmentation COCO minival CenterMask-VoVNetV2-99 (multi-scale) mask AP 42.5 # 37
Object Detection COCO minival CenterMask+VoVNetV2-57 (single-scale) box AP 44.6 # 78
APS 27.7 # 23
APM 48.3 # 21
Object Detection COCO test-dev CenterMask + X-101-32x8d (single-scale) box AP 44.6 # 110
AP50 63.4 # 98
AP75 48.4 # 86
APM 47.2 # 89
Hardware Burden None # 1
Operations per network pass None # 1
Instance Segmentation COCO test-dev CenterMask + ResNet-101-FPN mask AP 38.3 # 49
Instance Segmentation COCO test-dev CenterMask + VoVNetV2-99 (single-scale) mask AP 40.6 # 32
AP50 62.3 # 16
AP75 44.1 # 11
APS 20.1 # 18
APM 42.8 # 14
APL 57.0 # 12
Instance Segmentation COCO test-dev CenterMask + VoVNetV2-99 (multi-scale) AP50 66.2 # 8
AP75 47.4 # 8
APS 27.2 # 7
Instance Segmentation COCO test-dev CenterMask + VoVNetV2-57 (single-scale) AP50 60.8 # 22
APS 19.4 # 20
APM 41.7 # 21
Instance Segmentation COCO test-dev CenterMask + VoVNet99 mask AP 41.8 # 23
APS 24.4 # 8
APM 44.4 # 8
APL 54.3 # 14
Object Detection COCO test-dev CenterMask+VoVNetV2-99 (single-scale) box AP 45.8 # 102
AP50 64.5 # 81
APS 27.8 # 70
APM 48.3 # 78
APL 57.6 # 77
Hardware Burden None # 1
Operations per network pass None # 1
Instance Segmentation COCO test-dev CenterMask + X101-32x8d (single-scale) mask AP 39.6 # 42
AP50 61.2 # 20
AP75 42.9 # 16
APS 19.7 # 19
Object Detection COCO test-dev Centermask + ResNet101 AP50 61.6 # 116
AP75 46.9 # 99
Hardware Burden None # 1
Operations per network pass None # 1
Object Detection COCO test-dev CenterMask+VoVNet2-57 (single-scale) box AP 44.7 # 108
AP50 63.1 # 99
AP75 48.6 # 80
APS 27.1 # 77
APL 55.9 # 97
Hardware Burden None # 1
Operations per network pass None # 1
Object Detection COCO test-dev CenterMask-VoVNet99 (multi-scale) AP50 68.3 # 51
AP75 53.2 # 49
APS 32.4 # 36
APL 60.0 # 60
Hardware Burden None # 1
Operations per network pass None # 1
Real-time Instance Segmentation MSCOCO CenterMask-Lite (ResNet-50-FPN) mask AP 32.9 # 9
APS 12.9 # 4
APM 34.7 # 8
APL 48.7 # 9

Methods