laishenqi comments

Results 7 comments of


                                            laishenqi

模型解压报错

直接用torch.load加载就好，这个不是压缩包，只是随便加的后缀。

The last two FC layers don't need batch norm.

some details are not the same as original paper, because I do lots of experiments, so I just do as is the custom.

Why do you use conv and bn in SeModule?

Sorry, it's my fault to do so, I have retrain it, new model mbv3_small also get top-1 69.037, I have updated it.

Validation accuracy for large model low, mistakes in model

I add some tricks, some important tricks like warmup and cosine learning rate are really useful，besides, I use DALI bu Nvidia to load the model.

Validation accuracy for large model low, mistakes in model

I think the main cause is the dataloader, I will reproduce the model by dataloader in pytorch, instead of DALI.

Validation accuracy for mobilenetv3_small

I just use the DALI by Nvidia to read the dataset. For validation, I resize the short size to 256 and do center crop. Actually, I think it should be...

RandomIdentitySampler

这是为了保持数目一致，因为我们最后一个batch只采751个id，每个id共计4张图片，故而要len和这个统一，不然在训练中，就会出现iter才跑了一半，整个epoch直接结束跳入下一个epoch的情况，主要是为了保证统一。