How about the time cost of per image inferring for this project?

[lyw615]

Could you provide your time cost details when inferring? Just like which model used, average per image inferring time , average image size ,and Gpu type. If it's convenient for you, i will be appreciate a lot.

[willthefrog]

our model zoo doc has pretty detailed benchmark results, you can check it out.

[lyw615]

Face detection in this project faster than face detection in this link ?https://github.com/PaddlePaddle/models/tree/develop/PaddleCV/face_detection. I don't see any detailed benchmark results of the links. Where can i find it?

[willthefrog]

please refer to the face detection doc

[lyw615]

But i don't see anything about blazeFace and baceBoxes mentioned in the Readme.md here https://github.com/PaddlePaddle/models/tree/develop/PaddleCV/face_detection.

[lyw615]

My model is downloaded from Pyramidbox-v1-SSD 640x640 link, and with the pretrained model VGG. All the knowledges getting from the Readme.md of the model are above .

[willthefrog]

But i don't see anything about blazeFace and baceBoxes mentioned in the Readme.md here https://github.com/PaddlePaddle/models/tree/develop/PaddleCV/face_detection.

please see the link I provided.

[lyw615]

但是我真的是不知道我发你链接里那个人脸检测项目用的哪个architecture啊，我在人脸检测的时候都没有用过这个链接里的blazeface 和faceboxes 相关的yml文件https://github.com/PaddlePaddle/PaddleDetection/tree/release/0.1/configs/face_detection

[willthefrog]

pyramidbox is a different model, which is not included in paddledetection, since it's using VGG as backbnone, my guess is that the runtime speed will be much slower than blazeface. But if you really need detailed benchmark of pyramidbox, please submit an issue in models repo. Hope this clears it up for you.

[lyw615]

Does the prediction time include pre-processing of the image and post-processing time of the prediction result?

[willthefrog]

yes