U-Net shape

[zyazdani-92]

Hello, I trained U-Net with one-class batches with the size of (128,128), and get prediction results of the size of (128,128). I would like to use Smoothly-Blend-Image-Patches function to remove border artifacts in the prediction results of my trained U-Net model. My input images are 8-bit grayscale with a size of (799,799). When I use your function I have the following error:

WARNING:tensorflow:Model` was constructed with shape (None, 128, 128, 1) for input KerasTensor(type_spec=TensorSpec(shape=(None, 128, 128, 1), dtype=tf.float32, name='input_1'), name='input_1', description="created by layer 'input_1'"), but it was called on an input with incompatible shape (None, 128, 128, 3).

I applied #4 and #3 but still I have error. Which function should be changed to make this pipeline compatible with my U-Net model?
Thank you so much.

---

Update: I reshape my input image as follows:
input_img = io.imread('RatNeuron_5_666_Bw_2_5X.tif')

input_img = input_img.reshape((799,799,1))
And then could run the Smoothly-Blend-Image-Patches function
But the prediction result is extremely changed.

1. First image is my input image.
2. Second image is the prediction result without using Smoothly-Blend-Image-Patches function.
3. Third image is the prediction result with Smoothly-Blend-Image-Patches function.

[Mirzyaaliii]

Hello @zyazdani-92,

Were you able to find any solution using the code from this repository or any other code for smooth blending?

My scenario is similar to yours - grayscale image and output have only 1 class (if we do not consider background). I am also having an issue while using this code. So, without using smooth blending, I am getting good enough output but edge artifacts, and if I use this smooth blending code, I am getting just blank output (all black or 0 pixels).

Any solution??

[zyazdani-92]

Hi @Mirzyaaliii
I used the method mentioned here instead of using smooth blending:
Adding geodesic information and stochastic patch-wise image prediction for small dataset learning

[Mirzyaaliii]

I will check it out. Thanks for your reply, @zyazdani-92 !!