jiweibo / imagenet Goto Github PK

如题，请问这种情况正常吗？谢谢你

I clone the project and reexecute it with resnet50. However, it comes out only Prec@1 72.780 Prec@5 90.880. I only add "nn.DataParallel". Could you give me the final *.pth of resnet50？

Yours,
Sincerely.

你好，我使用你的代码训练resnet，最后的top1大概最高只能达到resnet18 68.5左右,resnet34 72.88左右。达不到你表格中的69.758%和73.314。请问resnet18、resnet34这两个训练的初始参数和学习率是怎样的？

您好，我在models文件夹下新建了mnet.py，设计了新的网络，然而执行main.py的model = mnet (pretrained=args.pretrained)时出现了nameerror。
请问该怎么解决啊？谢谢！

显卡内存不够，batchsize设置为200，前几个epoch TOP1和TOP5的准确率非常低，4个eopchs下来在验证集上TOP5只有0.005，是ImageNet数据集没弄好吗？还是本来就是这样呢？

There may be something wrong in top5 accuracy updating, which making the top5 accuracy is the same as the top1.
The code with problem(main.py) is as follow:

您好！请问从官网下载的几个tar压缩包，直接解压就可以使用吗？

Did you still have the hper-parameters for the different networks to achieve the top accuracy?

In helper.py Line 42: shutil.copyfile(filename, 'alex_model_best.pth')

Hi, jiweibo. Do you have a save for the training log of ResNet-34 on ImageNet? I need to plot it in my paper, thx!