YolactEdge: Real-time Instance Segmentation on the Edge
This work provides a significant speed improvement for real-time instance segmentation on resource-constrained edge devices, benefiting applications requiring on-device processing.
This paper presents YolactEdge, a real-time instance segmentation method designed for edge devices. It achieves up to 30.8 FPS on a Jetson AGX Xavier and 172.7 FPS on an RTX 2080 Ti, demonstrating a 3-5x speedup over existing real-time methods while maintaining competitive accuracy on YouTube VIS and MS COCO datasets.
We propose YolactEdge, the first competitive instance segmentation approach that runs on small edge devices at real-time speeds. Specifically, YolactEdge runs at up to 30.8 FPS on a Jetson AGX Xavier (and 172.7 FPS on an RTX 2080 Ti) with a ResNet-101 backbone on 550x550 resolution images. To achieve this, we make two improvements to the state-of-the-art image-based real-time method YOLACT: (1) applying TensorRT optimization while carefully trading off speed and accuracy, and (2) a novel feature warping module to exploit temporal redundancy in videos. Experiments on the YouTube VIS and MS COCO datasets demonstrate that YolactEdge produces a 3-5x speed up over existing real-time methods while producing competitive mask and box detection accuracy. We also conduct ablation studies to dissect our design choices and modules. Code and models are available at https://github.com/haotian-liu/yolact_edge.