| 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。
Citations
@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}
}
yangjun
dfa27afb39
提交PaddleDetection develop 分支 d56cf3f7c294a7138013dac21f87da4ea6bee829