here I met some questions , I hope to get your help

[alexliyang]

hi, thank you for your code.

when I use these codes, I met some problems:

1. when I test one image , although I expend_dims the image, I can not get the right predict, but ,when test a batch images , the predict is right

2. when I try to save ckpt to pb file, here reuse should be set True, but when set True, there will be some problems , when set False, _classification_layer's flatten0 = tf.reshape(global_avgpool, [shape[0], -1]) maybe wrong

what can I do for it?

[nnUyi]

for Q1:
self.input_train = tf.placeholder(tf.float32, [self.batchsize, self.input_height, self.input_width, self.input_channel], name='input_train')
self.input_test = tf.placeholder(tf.float32, [self.batchsize, self.input_height, self.input_width, self.input_channel], name='input_test')

Once you want to test only one image, you should rewrite this code in build_mode function of peleenet.py to be:
self.input_train = tf.placeholder(tf.float32, [None, self.input_height, self.input_width, self.input_channel], name='input_train')
self.input_test = tf.placeholder(tf.float32, [None, self.input_height, self.input_width, self.input_channel], name='input_test')

This requires no constraint in batchsize of training or testing phase.

for Q2:
Could you describe your problem in detail? I confuse about that. Thank you.