提交 3bb9ab77 编写于 作者: B baiyf 提交者: qingqing01

Add performance curve for face_detection (#1029)

* add AP figures
上级 90a40e8c
...@@ -160,4 +160,12 @@ python -u train.py --batch_size=16 --pretrained_model=vgg_ilsvrc_16_fc_reduced ...@@ -160,4 +160,12 @@ python -u train.py --batch_size=16 --pretrained_model=vgg_ilsvrc_16_fc_reduced
| 模型 | 预训练模型 | 训练数据 | 测试数据 | mAP | | 模型 | 预训练模型 | 训练数据 | 测试数据 | mAP |
|:------------------------:|:------------------:|:----------------:|:------------:|:----:| |:------------------------:|:------------------:|:----------------:|:------------:|:----:|
|[Pyramidbox-v1-SSD 640x640]() | [VGGNet](http://paddlemodels.bj.bcebos.com/vgg_ilsvrc_16_fc_reduced.tar.gz) | WIDER FACE train | WIDER FACE Val | 95.6%/ 94.6%/ 89.4% | |[Pyramidbox-v1-SSD 640x640]() | [VGGNet](http://paddlemodels.bj.bcebos.com/vgg_ilsvrc_16_fc_reduced.tar.gz) | WIDER FACE train | WIDER FACE Val | 95.6%/ 94.7%/ 89.3% |
#### 性能曲线
<p align="center">
<img src="images/wider_pr_cruve_int_easy_val.jpg" width="280" />
<img src="images/wider_pr_cruve_int_medium_val.jpg" width="280" />
<img src="images/wider_pr_cruve_int_hard_val.jpg" width="280" /></br>
WIDER FACE Easy/Medium/Hard set
</p>
...@@ -127,7 +127,6 @@ def detect_face(image, shrink): ...@@ -127,7 +127,6 @@ def detect_face(image, shrink):
det_ymax = image_shape[1] * detection[:, 5] / shrink det_ymax = image_shape[1] * detection[:, 5] / shrink
det = np.column_stack((det_xmin, det_ymin, det_xmax, det_ymax, det_conf)) det = np.column_stack((det_xmin, det_ymin, det_xmax, det_ymax, det_conf))
det = det[keep_index, :]
return det return det
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册