dcgan | The Simplest DCGAN Implementation | Machine Learning library

Note that the call to make_gen_block is actually calling self.make_gen_block. The self is important. You can see in the signature of __init__ that self is injected as the first argument. The method can be referenced because self has been passed into the __init__ method (so it is within the scope), and self is of type Generator, which has a make_gen_block method defined for it. The self instance of the class has already been constructed prior to the calling of the __init__ method.

When the class is instantiated, the __new__ method is called first, which constructs the instance and then the __init__ method is called, with the new instance injected (as self) into the method.

You can pass a list of tensors to the call function, so you could pass the labels. However, this is not in the logic of tensorflow/Keras training. In your example, the basic training routine is train_step. The output tensors are first calculated by the generator and discriminator call function, and then passed to the functions that calculate the losses. This is the standard way of doing things:

The training data has 60000 instances, if you split them into batches of size 256 you are left a smaller batch of size 60000 % 256 which is 96. Keras also assumes this as a batch if you dont drop it. So in train_step for this batch of size 96, the shape of real_output will be (96, 1) and the shape of fake_output will be (256, 1). As you set reduction to None in cross_entropy loss, the shape will be retained, so shape of real_loss will (96,) and shape of fake_loss will be (256,) then adding them will definitely result in an error.

You may solve this problem this way -

Yes the generator needs to be changed too. Greyscale has one channel and you need three.

So you need to change

Regarding the error message

RuntimeError: The size of tensor a (4) must match the size of tensor b (3) at non-singleton dimension 0

would lead to suggest that there's a problem with this call: sample = self.transform(sample)

Indeed, the issue is you are using a T.Normalize transform which only expects three channels (you specified a mean and std for three channels only, not four).

Using dset.ImageFolder, without explicitly defining the function that reads the image (the loader) results with your dataset using the default pil_loader:

I think the error means that it's trying to stack the images into batches, but the images are different sizes, i.e. the first image is size 3 x 64 x 85 and the second image is 3 x 64 x 80. You'll probably need to transform (resize) the images so that they're all the same shape.

For random number generation part, it is recommended to use the new API tf.random.Generator to generate random number.

Reference: https://www.tensorflow.org/guide/random_numbers

See below for an example of random number generation and swapping elements between tensors.

Example codes:

Did you try changing the version? if its still broken please share your logs and full code.