ValueError: Dimension 2 in both shapes must be equal, but are 320 and 64 for 'Assign_3' (op: 'Assign') with input shapes: [7,7,320,64], [7,7,64,3].

[sanalmgr]

Hi,

I am using the things (testing - by Command "python main.py sample_images/") as it is... but receiving the error, please help me how to resolve this error:

ValueError: Dimension 2 in both shapes must be equal, but are 320 and 64 for 'Assign_3' (op: 'Assign') with input shapes: [7,7,320,64], [7,7,64,3].

[marcellacornia]

Hi @sanalmgr, thanks for downloading our code.

Which Keras version are you using? This code is compatible with Keras 1.1.0 only. Also, you have to use Theano as backend. Please check in your keras.json file if the image_dim_ordering is set to "th".

[kfzyqin]

Same problem here. Keras version: 2.2.4

[kfzyqin]

Setting "image_data_format": "channels_first" solves the issue.

[1404020219]

ValueError: Dimension 0 in both shapes must be equal, but are 7 and 64. Shapes are [7,7,320,64] and [64,3,7,7]. for 'Assign' (op: 'Assign') with input shapes: [7,7,320,64], [64,3,7,7]. @ZhenyueQin After Setting "image_data_format": "channels_first" ,so sad==

[1404020219]

why?

[kfzyqin]

@1404020219 Sorry I could not solve more issues until I changed the Keras version to the author's one.

[oalvarezc]

Con la versión 2.2.4 keras, lo solucioné implementando en Conv2D data_format='channels_first'