Instance Segmentation on COCO 2017 (Accuracy and Latency)
35.9APsFastViT-SA12
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FastViT-SA12Backbone=FastViT-SA12, Framework=Mask R-CNN, Image Crop Size=512x512, Batch Size (GPU/CPU/Mobile)=32/16/12024.01 | 35.9 | 1.06 | 39.4 | 6.5 | 38.9 | 60.5 | 42.2 | 57.6 | 38.1 | — | — | — | |
| SHViT-S4Backbone=SHViT-S4, Framework=Mask R-CNN, Image Crop Size=512x512, Batch Size (GPU/CPU/Mobile)=32/16/12024.01 | 35.9 | 0.28 | 5 | 3.3 | 39 | 61.2 | 41.9 | 57.9 | 37.9 | — | — | — | |
| FastViT-SA12Head=Mask R-CNN, Latency (ms)=1.63, Resolution=512x512, Schedule=1x2024.11 | 35.9 | — | — | — | — | — | — | 57.6 | 38.1 | 38.9 | 60.5 | 42.2 | |
| SHVIT-S4Head=Mask R-CNN, Latency (ms)=0.52, Resolution=512x512, Schedule=1x2024.11 | 35.9 | — | — | — | — | — | — | 57.9 | 37.9 | 39 | 61.2 | 41.9 | |
| EfficientViM-M4Head=Mask R-CNN, Latency (ms)=0.45, Resolution=512x512, Schedule=1x2024.11 | 35.8 | — | — | — | — | — | — | 57.1 | 37.4 | 39.3 | 60.2 | 42.5 | |
| EfficientFormer-L1Backbone=EfficientFormer-L1, Framework=Mask R-CNN, Image Crop Size=512x512, Batch Size (GPU/CPU/Mobile)=32/16/12024.01 | 35.4 | 0.84 | 21 | 4.3 | 37.9 | 60.3 | 41 | 57.3 | 37.3 | — | — | — | |
| PoolFormer-S12Backbone=PoolFormer-S12, Framework=Mask R-CNN, Image Crop Size=512x512, Batch Size (GPU/CPU/Mobile)=32/16/12024.01 | 34.6 | 1.2 | 40.4 | 6.8 | 37.3 | 59 | 40.1 | 55.8 | 36.9 | — | — | — | |
| PoolFormer-S1Head=Mask R-CNN, Latency (ms)=1.49, Resolution=512x512, Schedule=1x2024.11 | 34.6 | — | — | — | — | — | — | 55.8 | 36.9 | 37.3 | 59 | 40.1 | |
| ResNet-50Backbone=ResNet-50, Framework=Mask R-CNN, Image Crop Size=512x512, Batch Size (GPU/CPU/Mobile)=32/16/12024.01 | 34.4 | 0.94 | 19 | 8.8 | 38 | 58.6 | 41.4 | 55.1 | 36.7 | — | — | — | |
| EfficientViT-M4Backbone=EfficientViT-M4, Framework=Mask R-CNN, Image Crop Size=512x512, Batch Size (GPU/CPU/Mobile)=32/16/12024.01 | 31 | 0.33 | 7.3 | 7.8 | 32.8 | 54.4 | 34.5 | 51.2 | 32.2 | — | — | — | |
| EfficientNet-B0Backbone=EfficientNet-B0, Framework=Mask R-CNN, Image Crop Size=512x512, Batch Size (GPU/CPU/Mobile)=32/16/12024.01 | 29.4 | 0.54 | 16.7 | 3.8 | 31.9 | 51 | 34.5 | 47.9 | 31.2 | — | — | — | |
| EfficientNet-B0Head=Mask R-CNN, Latency (ms)=0.95, Resolution=512x512, Schedule=1x2024.11 | 29.4 | — | — | — | — | — | — | 47.9 | 31.2 | 31.9 | 51 | 34.5 |