python - How to decide on the right target size of an input image for inception_v3? (predictions differ)

Question

I am running an Inception V3 model to make predictions. I am finding out that different target sizes for the input image lead to different predictions.

For the same image, I got different predictions. For (416, 416): tiger_cat (6.228156387805939), for (299, 299): Egyptian_cat (55.72159290313721), for (224, 224): lynx (48.244667053222656).

I thought that the input size should be divided by 32. Which is not the case for (299, 299). But, for another input size I get the error message:

WARNING:tensorflow:Model was constructed with shape (None, 299, 299, 3) for input Tensor("input_13:0", shape=(None, 299, 299, 3), dtype=float32), but it was called on an input with incompatible shape (None, 224, 224, 3).

How do I decide on the best target size?

question from:https://stackoverflow.com/questions/65898956/how-to-decide-on-the-right-target-size-of-an-input-image-for-inception-v3-pred

Answer 1

Waitting for answers