search

chenzhm01 / VGG16_slim

training vgg16 with single-gpu, multi-gpu, distribute by tensorflow-slim
3 stars 1 forks source link

VGG16_slim问题请教！ #1

Open 670373064 opened 6 years ago

670373064 commented 6 years ago

你好，很荣幸看到你的ＶＧＧ的finetune程序，在运行您的程序时，出了点问题，想向您请教！ 我是把您的４７类换成了２类，在运行下面的程序时，报错： net, endpoint = vgg.vgg_16(inputs, num_classes=None, is_training=is_training, dropout_keep_prob=dropout_keep_prob)

报错如下： Shape of a new variable (vgg_16/fc8/weights) must be fully defined, but instead was (1, 1, 4096, ?).

是 num_classes=None,　这个不能为Ｎone啊，但是我改成２又报错： ValueError: ('Convolution not supported for input with rank', 2)

我刚刚接触这个不久，，麻烦您帮忙解答一下，谢谢！！

chenzhm01 commented 6 years ago

我上传了nets文件夹，你再试试， 用num_classes=None