pretrained model with a test image: Invalid Argument and Incompatible Shapes

[zBabar]

I am trying to run pertained model with a test image. after settling lot of things still I got certain error like

InvalidArgumentError (see above for traceback): Incompatible shapes: [1,28,28,21] vs. [1,16,16,21]

[[Node: fcn_8s/add = Add[T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"](fcn_8s/pool4_fc/BiasAdd, fcn_8s/conv2d_transpose)]]

Do I need to change the shape of the image first? But for each test image it says different dimensions vs different dimensions

Moreover whats the purpose of num_classes in pertained model? and what should it be for a test image?

[Senyao-Han]

I believe it's a padding problem.
just go to slim/nets/vgg.py, change this line:
net = slim.conv2d(net, 4096, [7, 7], padding='VALID', scope='fc6')
to padding='SAME'. It will be okay.
The thing is, my result is not good. I've got low contrast images.