ValueError: Layer weight shape (3L, 3L, 640L, 64L) not compatible with provided weight shape (64, 3, 3, 3)

[xycim]

hi, when i run your codes, it doesnot work with hint("ValueError: Layer weight shape (3L, 3L, 640L, 64L) not compatible with provided weight shape (64, 3, 3, 3)"). Any advices for me?

[sino1Ping]

same error occurs to me! with win7 anaconda 4.2 python 2.7 and theano 0.9.0, any suggestion?

[baraldilorenzo]

Dear @xycim, @sino1Ping,

please see this issue: #2 and see if it works for you.

[sino1Ping]

hi @baraldilorenzo
Thank you for your reminding. Now the "image_data_format" and "backend" in json file configured properly. I modified "from keras import initializations" to "from keras import regularizers",due to the keras version 2.0.4 In win7 system, I run into new error : "Layer weight shape (3L, 3L, 3L, 64L) not compatible with provided weight shape (64, 3, 3, 3)" And in ubuntu system, the hint is "Layer weight shape (3, 3, 3, 64) not compatible with provided weight shape (64, 3, 3, 3)"

And when I set firt layer weights to None , the error hint is"Layer weight shape (3L, 3L, 64L, 64L) not compatible with provided weight shape (64, 64, 3, 3)"

and first two layers weights to None : "Layer weight shape (3L, 3L, 64L, 128L) not compatible with provided weight shape (128, 64, 3, 3)"

same issue with #1.but I'm sure I configured properly, and added backend.set_image_data_format("channels_first")" in the beginning of program. the same error still occurs

ps: I'm using theano 0.9.0 keras 2.0.4 both in win7 and ubuntu system

[marcellacornia]

Hi @sino1Ping, you have to use Keras 1.1.0. Unfortunately, this code is not compatible with Keras 2.

[sino1Ping]

Thanks! It works.

[kmamine]

Thanks! It works.

What did you do ?