Check if your CNN implementation is Correct

Is there a way to confirm that my own implementation of CNN is working correctly? I tired running the exact same model structure on Tensorflow but i am getting wildly different accuracy.

There is probably no automated way to do that, if that’s what you mean. It would require careful examination of the code and perhaps constructing more experiments. Or at least making sure you fully understand the meaning of the experiments you’ve constructed so far. If you are comparing hand written python code to a TF implementation, there are lots of other variables to worry about. E.g. random initialization is handled differently by TF, so you won’t get exactly the same results.