Running evaluation on images of different size

[burhanbvk]

Hello,

I am trying to create a service that evaluates images. To improve performance I am trying to use the same network on all images. I use inference.py as a starting point and I moved the net variable to the global scope using the following code:

img_ph = tf.placeholder(tf.float32, shape = [None, None, None, 3]) net = DeepLabResNetModel({'data': img_ph}, is_training=False, num_classes=NUM_CLASSES)

But I get the following error when I evaluate an image:

File "C:\Users\Administrator\AppData\Local\Programs\Python\Python36\lib\site-packages\tensorflow\python\framework\ops.py", line 5573, in _assert_same_graph original_item)) ValueError: Tensor("strided_slice:0", shape=(2,), dtype=int32) must be from the same graph as Tensor("fc_out/Conv2D:0", shape=(?, ?, ?, 27), dtype=float32).

Thanks for any pointers.

[DrSleep]

I am not sure where this is coming from

Tensor("fc_out/Conv2D:0", shape=(?, ?, ?, 27), dtype=float32)

Did you modify the network? If so, it is better if you provide more description; otherwise, I might not be able to help at all