Reproduce WIDER score

[dereyly]

Hello I try to reproduce MTCNNv1 WIDER score on validation data I have not exactly same resault as in paper or mat files eval/plot/baselines/Val/setting_int/multitask-cascade-cnn/wider_pr_info_multitask-cascade-cnn_easy_val.mat I set parameters: threshold=[0.5 0.5 0.3]; factor=0.79; And have: 0.834 0.810 0.624 vs 0.848 0.825 0.598 Is difference more then 1% mAP significantly?

[LinnaWang76]

Hello Could you solve it?i have the same problems! But I got worse results. The same parameters: And have: 0.517 0.329 0.218 I have no idea to reproduce the WIDER score

[zzzzzz0407]

hello,i would like to know the speed of detect,in paper it just said 99fps ,but my speed is juast 2fps,what about you? my gpu is tatian x

[dereyly]

Hi I set minsize=30 %minimum face size And it runs 20fps-80fps

[zzzzzz0407]

@dereyly Hi, i just set the minimum=30, and other parameter is just like you, and test it on the wider face train set, only to find that the speed is just about 5fps, can you figure what's wrong with it?And my gpu is nvidia k80.

[dereyly]

@zHanami sorry for disinformation I run test on CelebA with min_face=60 and it is 20fps-80fps Test on wider with min_face=30 is 5.5fps as your resault Cascade like MTCNN potential more faster then classic detection like SSD, R-FCN and others And you need production code mostly cuda and some optimal C++ to get fps like 99