Results 28 comments of SyGoing

hello, have you solved this question? my architecture is the same as you

maybe the real architecture is only one L1 loss and L2 loss. I am trying it again !

yean ,like your architecture(4 stages L2 loss and 2 stages L1 loss, I use the vgg19 pretrained model to initialize the pose model , and the loss is changing smaller...

Hello , can you share the ultra-high precision model you trained? my email: [email protected] thanks!

and the resnet18 cannot be download,could you please put it on the baiduyun, thanks so much

when converting the CenterNet-FPN_R50_1x [03/20 10:56:30 d2.data.common]: Serializing 5000 elements to byte tensors and concatenating them all ... [03/20 10:56:30 d2.data.common]: Serialized dataset takes 19.10 MiB Traceback (most recent call...

@gneworld 你好,建议你好好看看论文,另外这个RF_center_Xs_mat[j]这个是感受野中心位置的X坐标,constant是该层的stride,box_map.channel(0)[j] 是左上角预测的相对感受野中心的偏置,这个你要是有疑惑建议手动计算一下感受野及其感受野中心位置 可以参考一下官方issue:https://github.com/YonghaoHe/A-Light-and-Fast-Face-Detector-for-Edge-Devices/issues/28

你好,请你参看readme 转换模型就可以了 ------------------ 原始邮件 ------------------ 发件人: "hanfanggithub"; 发送时间: 2019年10月8日(星期二) 下午2:16 收件人: "SyGoing/LFFD-with-ncnn"; 抄送: "Subscribed"; 主题: Re: [SyGoing/LFFD-with-ncnn] ncnn error (#5) 我用v2版本人脸检测测试的时候显示的错误是slice_axis不支持,我想问下你们用的ncnn是哪一个版本 这个问题你解决了吗 — You are receiving this because you are...

@dhdljd,你好,关于运行时间,你可以设置输入320x240(如果应用场景视场较小,人脸较大),甚至160x140,或许会快很多。另外关于windows下编译,我建议你自己配置一个vs2019工程直接跑,如果非要用cmake 请显式的指定opencv include 和lib,不要find_package(OpenCV REQUIRED),实际运行的时候请在exe所在目录加入MNN.dll和opencv_worldXXX.dll.

你好,这个转换我这边是没问题,不知道具体你是什么问题,建议你按照详细步骤进行,有一点不对都会出错的。