After training the model (for a short time) I see an accuracy of 80%. However, when I test it, the result is as shown for all test images. Can someone help with this? Would be great if someone could share the pre-trained weights.

When I train a model with square input size, all goes well. However, when I train a model with an input layer of, for example, 512x768, the training loss increases to up to 97% but the output is almost completely empty.

Anybody succeeded using UNet with a non-square input size?

i) in your model.py input_size (256,256,1) => (256,256,3) ii) trainGenerator function in your data.py image_color_mode ="grayscale"=>"rgb" iii) testGenerator function in your data.py
comment out "img = np.reshape(img,img.shape+(1,)) if (not flag_multi_class) else img"

Originally posted by @teamo429 in https://github.com/zhixuhao/unet/issues/59#issuecomment-439797032

I followed the above steps but I get the following error: ValueError: Error when checking target: expected conv2d_24 to have shape (256, 256, 1) but got array with shape (256, 256, 3)

Hi, I'm trying to do a color segmentation (I have grayscale image, and their labeled image with colors).

I changed mask_color_mode to "rgb", and num_class to 3 (because I have 3 differents colors) in data.py, and I changed input_size from 256,256,1 to 256,256,3 in model.py, but I still have errors..

"expected input_1 to have shape (None,256,256,3) but got array with shape (2,256,256,1). I rune the debugger and, mask.shape gave me : (2,256,256,3) and omg.shape gave me : (2,256,256,1)

Any idea ? Thanks in advance :)

I have trained my model The weights are now stored in unet_membrane.hdf5

I want to use these previously-stored weights for testing it on images now.

How to do it?

Hello, in the trainGenerator, you use one image_datagen and one mask_datagen, but they load data from different directory, how could you say ---the label of image 1 is actually the label of image 1?

Hi There! By default input size of image is 256x256. Now my images resolution is 1364x659 and when i compile the main file i get this error ValueError: A Concatenate layer requires inputs with matching shapes except for the concat axis. Got inputs shapes: [(None, 164, 341, 256), (None, 164, 340, 256)]

You have achieved training accuracy of 97% in 5 epoch but calculation of test accuracy is not provided.Could you provide information on test accuracy ?

I am running the mentioned notebook. All is fine until step 5.2 when I get the following output: Model input size: 512x512 Number of dataset found in the folder: 30

100% 30/30 [00:02<00:00, 13.98it/s]

Running QC on: 8.png

FileNotFoundError Traceback (most recent call last)

in () 52 print('Running QC on: '+filename) 53 test_input = io.imread(os.path.join(Source_QC_folder, filename), as_gray=True) ---> 54 test_ground_truth_image = io.imread(os.path.join(Target_QC_folder, filename), as_gray=True) 55 56 (threshold_list, iou_scores_per_threshold) = getIoUvsThreshold(os.path.join(prediction_QC_folder, prediction_prefix+filename), os.path.join(Target_QC_folder, filename))

6 frames

/usr/local/lib/python3.7/dist-packages/imageio/core/request.py in _parse_uri(self, uri) 381 # Reading: check that the file exists (but is allowed a dir) 382 if not os.path.exists(fn): --> 383 raise FileNotFoundError("No such file: '%s'" % fn) 384 else: 385 # Writing: check that the directory to write to does exist

FileNotFoundError: No such file: '/content/gdrive/MyDrive/Data/membrane/membrane-model/Quality Control/Prediction/8.png'

I have used the following code to transform the pytorch model into onnx and found that the output of onnx is different from the original pytorch model.

unet_model = UNet(n_channels=3, n_classes=1, bilinear=True)

model_weight = './checkpoints/CP_epoch100.pth'
unet_model.load_state_dict(torch.load(model_weight))


# Input to the model
batch_size = 1

x = torch.randn(batch_size, 3, 160, 128, requires_grad=True)

torch_out = unet_model(x)

# Export the model
torch.onnx.export(unet_model,  # model being run
                  x,  # model input (or a tuple for multiple inputs)
                  "./onnx/points_unet.onnx",  # where to save the model (can be a file or file-like object)
                  export_params=True,  # store the tra-ined parameter weights inside the model file
                  opset_version=11,  # the ONNX version to export the model to
                  do_constant_folding=True,  # whether to execute constant folding for optimization
                  input_names=['input'],  # the model's input names
                  output_names=['output'],  # the model's output names
                  dynamic_axes={'input': {0: 'batch_size'},  # variable length axes
                                'output': {0: 'batch_size'}})

I have a problem when run code in Colab. Can you guys please tell me what's happening ? Thanks!

WARNING:tensorflow:From :11: Model.fit_generator (from tensorflow.python.keras.engine.training) is deprecated and will be removed in a future version. Instructions for updating: Please use Model.fit, which supports generators. Found 0 images belonging to 1 classes. Found 0 images belonging to 1 classes.

ValueError Traceback (most recent call last) in () 9 model = unet() 10 model_checkpoint = ModelCheckpoint('unet_membrane.hdf5', monitor='loss',verbose=1, save_best_only=True) ---> 11 model.fit_generator(myGene,steps_per_epoch=2000,epochs=5,callbacks=[model_checkpoint])

10 frames <array_function internals> in amax(*args, **kwargs)

/usr/local/lib/python3.6/dist-packages/numpy/core/fromnumeric.py in _wrapreduction(obj, ufunc, method, axis, dtype, out, **kwargs) 88 return reduction(axis=axis, out=out, **passkwargs) 89 ---> 90 return ufunc.reduce(obj, axis, dtype, out, **passkwargs) 91 92

ValueError: zero-size array to reduction operation maximum which has no identity

• 默认数据集，训练5轮
• 预测发现全部是灰图，没有结果

看loss比较大，loss: 0.5252 - accuracy: 0.7814，就是没有模型收敛。 修改loss，lr，Adam(lr = 1e-5)，steps_per_epoch=2000, epochs=20，batch_size=32。 loss缓慢下降，正常收敛，预测正常 关键是batch_size和lr需要微调，多调整几次看loss有没有下降趋势，没有的话中断训练，继续修改

2022-09-13 05:10:38.208563: I tensorflow/stream_executor/cuda/cuda_dnn.cc:384] Loaded cuDNN version 8201 2000/2000 [==============================] - ETA: 0s - loss: 0.2607 - accuracy: 0.8837
Epoch 1: loss improved from inf to 0.26072, saving model to unet_membrane.hdf5 2000/2000 [==============================] - 858s 424ms/step - loss: 0.2607 - accuracy: 0.8837 Epoch 2/20 2000/2000 [==============================] - ETA: 0s - loss: 0.1718 - accuracy: 0.9249
Epoch 2: loss improved from 0.26072 to 0.17182, saving model to unet_membrane.hdf5 2000/2000 [==============================] - 849s 425ms/step - loss: 0.1718 - accuracy: 0.9249 Epoch 3/20 2000/2000 [==============================] - ETA: 0s - loss: 0.1499 - accuracy: 0.9342
Epoch 3: loss improved from 0.17182 to 0.14994, saving model to unet_membrane.hdf5 2000/2000 [==============================] - 850s 425ms/step - loss: 0.1499 - accuracy: 0.9342 Epoch 4/20 2000/2000 [==============================] - ETA: 0s - loss: 0.1355 - accuracy: 0.9406
Epoch 4: loss improved from 0.14994 to 0.13549, saving model to unet_membrane.hdf5 2000/2000 [==============================] - 851s 426ms/step - loss: 0.1355 - accuracy: 0.9406 Epoch 5/20 2000/2000 [==============================] - ETA: 0s - loss: 0.1243 - accuracy: 0.9455
Epoch 5: loss improved from 0.13549 to 0.12432, saving model to unet_membrane.hdf5 2000/2000 [==============================] - 841s 421ms/step - loss: 0.1243 - accuracy: 0.9455 Epoch 6/20 2000/2000 [==============================] - ETA: 0s - loss: 0.1154 - accuracy: 0.9495
Epoch 6: loss improved from 0.12432 to 0.11543, saving model to unet_membrane.hdf5 2000/2000 [==============================] - 841s 420ms/step - loss: 0.1154 - accuracy: 0.9495 Epoch 7/20 2000/2000 [==============================] - ETA: 0s - loss: 0.1082 - accuracy: 0.9527
Epoch 7: loss improved from 0.11543 to 0.10821, saving model to unet_membrane.hdf5 2000/2000 [==============================] - 839s 419ms/step - loss: 0.1082 - accuracy: 0.9527 Epoch 8/20 2000/2000 [==============================] - ETA: 0s - loss: 0.1022 - accuracy: 0.9554
Epoch 8: loss improved from 0.10821 to 0.10224, saving model to unet_membrane.hdf5 2000/2000 [==============================] - 840s 420ms/step - loss: 0.1022 - accuracy: 0.9554 Epoch 9/20 2000/2000 [==============================] - ETA: 0s - loss: 0.0972 - accuracy: 0.9576
Epoch 9: loss improved from 0.10224 to 0.09725, saving model to unet_membrane.hdf5 2000/2000 [==============================] - 843s 422ms/step - loss: 0.0972 - accuracy: 0.9576

2022-07-14 13:42:31.409488: I tensorflow/stream_executor/cuda/cuda_dnn.cc:384] Loaded cuDNN version 8400

Process finished with exit code -1073740791 (0xC0000409)

How can I fix this mistake?I would appreciate you very much!

Thank you for your code, Please allows me ask, does your code use the aug dataset? We ran finishly the prepare.ipynb file. At this time, we had the agument data in the "aug" folder. And in your main.py, the link data to the train folder ( label and image) not aug folder. I though, your main.py only use the original training dataset. Is it right? thanks. Thanks.

committed change :1) Uptaded model.py and decreased the learning rate . I got better result and the accuracy increased to 0.7944 2) Fixed the keras dependency error .

Hi,

Can someone help me with code customization for input size 1920x1080 image sizes. Working on a passion project right now, this would be really helpful. I do not wish to pad dummy background on image to make it 1920X1920(not losing any data in conversion to square shape, also would heavily increase the load on the GPU). Can someone help me with the Code customization required for the same?

Thanks a ton.