网络网络 | 输入尺寸 | 图片数/GPU | Epochs | 模型推理耗时(ms) | mAPval 0.5:0.95 |
Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
---|---|---|---|---|---|---|---|---|---|
YOLOF-R_50_C5 (paper) | 800x1333 | 4 | 12 | - | 37.7 | - | - | - | - |
YOLOF-R_50_C5 | 800x1333 | 4 | 12 | - | 38.1 | 44.16 | 241.64 | 下载链接 | 配置文件 |
注意:
- YOLOF模型训练过程中默认使用8 GPUs进行混合精度训练,总batch_size默认为32。
@inproceedings{chen2021you,
title={You Only Look One-level Feature},
author={Chen, Qiang and Wang, Yingming and Yang, Tong and Zhang, Xiangyu and Cheng, Jian and Sun, Jian},
booktitle={IEEE Conference on Computer Vision and Pattern Recognition},
year={2021}
}