search

GeorgeSeif / Semantic-Segmentation-Suite

Semantic Segmentation Suite in TensorFlow. Implement, train, and test new Semantic Segmentation models easily!
2.51k stars 880 forks source link

Error with image size #59

Closed bruno-krinski closed 6 years ago

bruno-krinski commented 6 years ago

Hi, I had this error: InvalidArgumentError (see above for traceback): logits and labels must be same size: logits_size=[36864,2] labels_size=[40000,2]. I think this is why my masks are in gray scale (I have only two classes). Any suggestions that I can run without this error?

sanja7s commented 6 years ago

I also have similar issues with several models (but not with DeepLab-V3, for instance).

Running DeepLab-V3 with 1000x1000 images (grayscale or 3 channel, RGB) works fine.

With the same data and using GCN I get error:

InvalidArgumentError (see above for traceback): Incompatible shapes: [1,21,63,63] vs. [1,21,64,64]

and using RefineNet-Res152 the error is:

InvalidArgumentError (see above for traceback): Incompatible shapes: [1,64,64,256] vs. [1,63,63,256]

Could you point us please if we should shape our data differently and how to adapt to different models?

Thanks!

GeorgeSeif commented 6 years ago

Please see issue #27