search

MaybeShewill-CV / attentive-gan-derainnet

Unofficial tensorflow implemention of "Attentive Generative Adversarial Network for Raindrop Removal from A Single Image (CVPR 2018) " model https://maybeshewill-cv.github.io/attentive-gan-derainnet/
MIT License
262 stars 83 forks source link

网络训练输入尺寸 #36

Closed Penn000 closed 5 years ago

Penn000 commented 5 years ago

您好！请问训练时的输入如果不是256x376，除了config/global_config.py还需要修改哪些文件呢？还是说输入的训练必须是256x376的，否则reshape到该尺寸？谢谢！

MaybeShewill-CV commented 5 years ago

@Penn000 只需要修改global config文件 如果出现维度不匹配的情况你需要根据情况修改输入图像的维度:)

Penn000 commented 5 years ago

谢谢您的回复:D global_config.py里的__C.TRAIN.IMG_HEIGHT等参数在train_model.py, derain_drop_net.py等文件中都没有用到，貌似对训练并不产生影响。之前我先train了256x376的数据集，然后在原仓库中再train别的大小的数据集就出现了InvalidArgumentError，现在我把代码拷到别的地方就能train起来了，应该是某个我没发现的缓存文件导致的。谢谢您啦~