search

PatrickHua / SimSiam

A pytorch implementation for paper 'Exploring Simple Siamese Representation Learning'
MIT License
809 stars 135 forks source link

Normalization for different datasets not implemented? #29

Open hanseungwook opened 3 years ago

hanseungwook commented 3 years ago

First of all, thanks for providing this pytorch implementation.

If we look into the augmentations for each of the models (SimSiam, BYOL, etc), it seems that it is using the ImageNet dataset's mean & std dev, regardless of whether you're training on CIFAR10 or CIFAR100 or others. (https://github.com/PatrickHua/SimSiam/blob/01d7e7811ac7b864bf8adccc8005208878208994/augmentations/simsiam_aug.py#L8)

Is my understanding correct and should this implementation be corrected?

hanseungwook commented 3 years ago

Also, the image size defaults to 224x224 (even for CIFAR10, CIFAR100, where the image sizes are 32x32). This is also a bug, correct?

BlankOtis commented 5 months ago

你好，我想问一下，这个模型在自己的数据集上训练会受类别与cafir-10不同而训练失败吗？