Chengpeng Chen comments

Results 9 comments of


                                            Chengpeng Chen

Ghost-ResNet-50 结构flops和parameters

我试过，但BN参数的差别微乎其微。

Ghost-ResNet-50 结构flops和parameters

> 看起来结构是对的，可能是BN之类的参数统计差异？请问可以分享一下Ghost-ResNet-50的代码或者ckpt吗？我想做一些对比实验，谢谢！

Ghost-ResNet-50 结构flops和parameters

感谢作者！已经run起来了：）不过这个模型测出来的params 跟我上面测的差不多。。

Ghost-ResNet-50 结构flops和parameters

> > 感谢作者！已经run起来了：）不过这个模型测出来的params 跟我上面测的差不多。。 > > 大佬，你是怎么运行的计算flops的代码，我就在iamhankai大佬上传的代码后面加上了 if **name** == '**main**': > > ``` > img_shape = (3,32,32) > > resnet18 = resnet18() > stat(resnet18, img_shape) # https://github.com/Swall0w/torchstat >...

Ghost-ResNet-50 结构flops和parameters

> 您好，您用的gost-resnet有预训练模型吗或者利用ghost_module修改后的resnent初始权重怎么设计的呢？我用的是pytorch自带的初始化，没有用预训练模型

Support RepGhostNetV2 conv s/m/l, based on MobileNetV4, without weights though

> @ChengpengChen thanks, once I'm done/satisified with the remaining mobilenetv4 model trainings I'll see how these fair with comparable hparams. > > FYI, if there is any reparameterization scheme to...

Support RepGhostNetV2 conv s/m/l, based on MobileNetV4, without weights though

> @ChengpengChen started training a large @ 384x384 with exact hparams of the mnv4 conv-large I trained. Hi, @rwightman Is there any update on the training results?

Support RepGhostNetV2 conv s/m/l, based on MobileNetV4, without weights though

> @ChengpengChen got sidelined by a bit of time off the other week... so I did finish training the large, but it's ~ .5% or so behind the mnv4 equivalent,...

Hi~Great work there! Is INT8 quantization all good?

Nice question! As we all know, a re-parameterized model is less quantization-friendly than a regular one. In our work, we have not evaluated the quantization performance of RepGhostNet yet. It...