测试速度-420ms/张 about centerface

[ywlife]

不正常，图像的宽高分别是多少？

[zhanglaplace]

@ywlife 摄像头默认读取的是720*1280的。后续测试输入了512X288的图片测试时间是 100ms。
640X480的时间是 200ms ,这个时间正常吗～

[ywlife]

晚回复是因为等机器跑测试数据。机器有限，cpu又一直在生成数据。之前测试：[email protected]/VGA视频可以30fps。下面是一直数据：（cpu还是有在生成数据，所以慢了点）
cpu times = 0:00:00.064898
cpu times = 0:00:00.061804
cpu times = 0:00:00.062592
cpu times = 0:00:00.075271
cpu times = 0:00:00.059055
cpu times = 0:00:00.069160
cpu times = 0:00:00.065846
cpu times = 0:00:00.067331
cpu times = 0:00:00.065196

另外，nms是python的，换成C++的才会好用。

[jiagh2010]

There are two questions: first, I also calculate the average time of different resolution pictures is more than 100 milliseconds; second, how to use GPU acceleration? @ywlife

[ywlife]

1. May be difference from test computer.
2. You can use caffe2 or other frameworks to load the onnx, and then run on GPU.

[jiagh2010]

Other frameworks can also call the onnx model？？？@ywlife

[zhanglaplace]

Other frameworks can also call the onnx model？？？@ywlife

@jiagh2010 Onnx to caffe2 . https://github.com/qfgaohao/pytorch-ssd/blob/master/convert_to_caffe2_models.py

[jiagh2010]

@jiagh2010 Onnx to caffe2 . https://github.com/qfgaohao/pytorch-ssd/blob/master/convert_to_caffe2_models.py I will convert the model to the Pb format of caffe2, but I will not call it. Can you give me an example?@ywlife @zhanglaplace

[zhanglaplace]

@jiagh2010 Onnx to caffe2 . https://github.com/qfgaohao/pytorch-ssd/blob/master/convert_to_caffe2_models.py I will convert the model to the Pb format of caffe2, but I will not call it. Can you give me an example?@ywlife @zhanglaplace
try it :
https://github.com/qfgaohao/pytorch-ssd/blob/master/convert_to_caffe2_models.py .

[lufi1]

Hi,

Im facing the same issue. Any idea about how I could deal with it ?
I m on macos and trying to reproduce your results