I have created a custom generator that yields a tuple of corrupted_image and original_image:

def custom_image_generator(files, data_instances, batch_size = 64):

    iter = 0
    num_batches = data_instances / batch_size
    while True:

        iter = iter % num_batches

        batch_input = []
        proc_batch = []

        start = random.randint(0, data_instances - batch_size)
        for index in range(start, start   batch_size):
            original_image = Image.open(files[index])
            corrupted_image = draw_square_on_image(files[index])
            batch_input.append(original_image)
            proc_batch.append(corrupted_image)

        corrupted_images_batch = np.array(proc_batch)
        original_images_batch = np.array(batch_input)

        iter = iter   1

        yield (corrupted_images_batch, original_images_batch)

If I call this generator like this:

(corrupted_images_batch, orig_images_batch) = next(test_generator)

It returns the expected output i.e 64 images in both the batches. For training my model I need to iterate over the entire batch.

But if I try doing something like:

for (corrupted_images_batch, orig_images_batch) in next(test_generator):
    print(corrupted_images_batch)

I get an error: ValueError: too many values to unpack (expected 2).

CodePudding user response：

As demonstrated by (corrupted_images_batch, orig_images_batch) = next(test_generator), next(test_generator) is a 2-tuple, so you can't loop over it unpacking each element into a 2-tuple.

What you're looking for is:

for (corrupted_images_batch, orig_images_batch) in test_generator:
    print(corrupted_images_batch)

This way you're looping over the generator, not just the next element generated.