English | 简体中文
These semantic segmentation models are designed for mobile and edge devices.
MobileSeg models adopt encoder-decoder architecture and use lightweight models as encoder.
Sandler, Mark, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. "Mobilenetv2: Inverted residuals and linear bottlenecks." In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 4510-4520. 2018.
Howard, Andrew, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang et al. "Searching for mobilenetv3." In Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 1314-1324. 2019.
Ma, Ningning, Xiangyu Zhang, Hai-Tao Zheng, and Jian Sun. "Shufflenet v2: Practical guidelines for efficient cnn architecture design." In Proceedings of the European conference on computer vision (ECCV), pp. 116-131. 2018.
Yu, Changqian, Bin Xiao, Changxin Gao, Lu Yuan, Lei Zhang, Nong Sang, and Jingdong Wang. "Lite-hrnet: A lightweight high-resolution network." In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 10440-10450. 2021.
Han, Kai, Yunhe Wang, Qi Tian, Jianyuan Guo, Chunjing Xu, and Chang Xu. "Ghostnet: More features from cheap operations." In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 1580-1589. 2020.
| Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) | Links |
|---|---|---|---|---|---|---|---|
| MobileSeg | MobileNetV2 | 1024x512 | 80000 | 73.94% | 74.32% | 75.38% | model | log | vdl |
| MobileSeg | MobileNetV3_large_x1_0 | 1024x512 | 80000 | 73.47% | 73.72% | 74.72% | model | log | vdl |
| MobileSeg | Lite_HRNet_18 | 1024x512 | 80000 | 70.75% | 71.62% | 72.53% | model | log | vdl |
| MobileSeg | ShuffleNetV2_x1_0 | 1024x512 | 80000 | 69.46% | 70.00% | 70.90% | model | log | vdl |
| MobileSeg | GhostNet_x1_0 | 1024x512 | 80000 | 71.88% | 72.22% | 73.08% | model | log | vdl |
| Model | Backbone | V100 TRT Inference Speed(FPS) | Snapdragon 855 Inference Speed(FPS) |
|---|---|---|---|
| MobileSeg | MobileNetV2 | 67.57 | 27.01 |
| MobileSeg | MobileNetV3_large_x1_0 | 67.39 | 32.90 |
| MobileSeg | Lite_HRNet_18 | 10.5 | 13.05 |
| MobileSeg | ShuffleNetV2_x1_0 | 37.09 | 39.61 |
| MobileSeg | GhostNet_x1_0 | 35.58 | 38.74 |
Note that:
- Test the inference speed on Nvidia GPU V100: use PaddleInference Python API, enable TensorRT, the data type is FP32, the dimension of input is 1x3x1024x2048.
- Test the inference speed on Snapdragon 855: use PaddleLite CPP API, 1 thread, the dimension of input is 1x3x256x256.