TF_CenterNet icon indicating copy to clipboard operation
TF_CenterNet copied to clipboard

Performance and backbone

Open phucnhs opened this issue 4 years ago • 6 comments

Thanks for source code ,

  • Can you tell me the performance on coco or voc on backbone mobilenetv2?
  • I want to add backbone mobilenetv3, I have to do it.
  • I think the backbone DLA is quite good, why don't you add it in this code ! Thanks you

phucnhs avatar Mar 28 '20 15:03 phucnhs

我也想知道不同backbone map多少 ,我自己用mobilenetv3训练过

dl8207531 avatar Apr 09 '20 02:04 dl8207531

我也想知道不同backbone map多少 ,我自己用mobilenetv3训练过

请问你有跑backbone是resnet18的吗?mAP是多少呢?

luqioayun avatar Jul 03 '20 03:07 luqioayun

voc map mobilenetv3 384x384 69+ 512x512 70+ resnet18 384x384 70+ 512x512 没测 @luqioayun

dl8207531 avatar Jul 03 '20 06:07 dl8207531

voc map mobilenetv3 384x384 69+ 512x512 70+ resnet18 384x384 70+ 512x512 没测 @luqioayun

好的,谢谢! 我刚好搞了512x512的,只有60+ 。。。。。。

luqioayun avatar Jul 03 '20 06:07 luqioayun

voc map mobilenetv3 384x384 69+ 512x512 70+ resnet18 384x384 70+ 512x512 没测 @luqioayun

image 1、您好,请问在resnet18 384x384 70+mAP这种情况下您的一些参数设置是什么呢?比如说learning rate的衰减方式、batchsize的大小设置~ image 这个是我目前的参数设置~

luqioayun avatar Jul 05 '20 07:07 luqioayun

lr=0.000125 batchsize=24 70000, 110000, 140000 对应步数除10 你根据你的batchsize 按比例设置 我认为最关键的是在imagenet预训练的模型本身不能太差

dl8207531 avatar Jul 05 '20 13:07 dl8207531