Model doesn't give high probability though it's accuracy is 85%

Hello dears, I’ve built my first deep learning model using CNN. My issue is with too low probabilities I get from the model though the model accuracy is 85% roughly.
I’ve tried many other photos but nothing helped out. May you help me please where the issue is?
Here is the project notebook
Deep Learning Model Classifier

@ptrblck May you help me please?


Can you report probabilities for a sample test/train image?
And also, Could you please explain why are you using log(softmax)? For comparison purposes or in your case, softmax is a more interpretable choice. Softmax maps input to probability space in range of [0, 1] so applying log on top of it, will lead to a all negative numbers but sum of them won’t be 1.


using softmax instead of log(softmax) giving constant accuracy

That is strange.

But about probabilities. Could you please explain how probabilities are not high enough? I mean, what is your comparison metric or let’s say for a image x, what do you expect and what are probs currently you obtaining?