'Conv2DBackpropInput' returned too few gradients. Expected 3 but received 2

[Shportaluk]

After added "Conv2DTranspose" to Generator i have exception in gradient

But if delete "Conv2DTranspose" i have not exception. Google not know how to fix this

[AsakusaRinne]

The problem is resolved by #980. Since the complete code is not given here, I use the following code to test it:

    [TestMethod]
    public void Test()
    {
        var model = keras.Sequential(name: "GENERATOR", layers:new List<ILayer>()
        {
            keras.layers.InputLayer(20),
            keras.layers.Dense(7 * 7 * 256),
            keras.layers.BatchNormalization(),
            keras.layers.LeakyReLU(),
            keras.layers.Reshape((7, 7, 256)),
            keras.layers.Conv2DTranspose(128, (5, 5), strides: (1, 1), output_padding: "same", use_bias: false, activation: "relu"),
            keras.layers.BatchNormalization(),
            keras.layers.LeakyReLU(),
            keras.layers.Conv2DTranspose(64, (5, 5), strides: (2, 2), output_padding: "same", use_bias: false, activation: "relu"),
            keras.layers.BatchNormalization(),
            keras.layers.LeakyReLU(),
            keras.layers.Conv2DTranspose(1, (5, 5), strides: (2, 2), output_padding: "same", use_bias: false, activation: "tanh"), // 28 28 1
        });
        
        model.summary();

        model.compile(new Adam(0.001f), new LossesApi().SparseCategoricalCrossentropy(), new string[] { "accuracy" });

        var num_epochs = 1;
        var batch_size = 8;

        var dataset = new RandomDataSet(new Shape(16, 20), 16);

        model.fit(dataset.Data, dataset.Labels, batch_size, num_epochs);
    }

    public class RandomDataSet : DataSetBase
    {
        private Shape _shape;

        public RandomDataSet(Shape shape, int count)
        {
            Data = np.random.normal(0, 2, shape);
            Labels = np.random.uniform(0, 1, (count, 28, 28, 1));
        }
    }

The PR has not been reviewed and merged. If time is limited for your project, please clone the branch of #980 and have a try.