question about learning rate

[gozian2811]

Mr Su,
I ran your code of mvcnn, and I've found a problem in the code. The training epoch is split into 3 stages, while theoretically the learningRates of every layer differ in different stage. However, the learningRates cannot be modified when stage changes, and only the last layer of the net may change during the whole training process.

[wangchu1]

I have found the exact issue as well. I am curious to know did Mr. Su reported their accuracy when only the last layer is learned? Or he had a working version of the code.