RuntimeError: stack expects each tensor to be equal size, but got [5898240] at entry 0 and [17694720] at entry 1 - MeanIoU

[CalinLucian]

Following these 2 tutorials (mixing data fetching + model from DVL3+ (keras_cv)

Tutorial 1 : https://keras.io/examples/vision/oxford_pets_image_segmentation/ Tutorial 2 : https://keras.io/examples/vision/deeplabv3_plus/

In tutorial 2, mean_iou computation proceeds with no issues.

I attach my code below where I did the processing and the training.

I get that 5898240 * 3(nb_of_classes) == 17694720 , but I do not understand what to modify at my code in order for it to work. If I delete the keras.metrics.MeanIoU(num_classes=3), the training proceeds successfully.

keras=3.0.1
keras-cv==0.7.2
torch==2.1.1+cu118

IMAGE_WIDTH = 640
IMAGE_HEIGHT = 384
BATCH_SIZE = 24

#berkley dataset, current lane, alternate lane, background
NUM_CLASSES = 3

DATA_DIR = "./drivable_maps"
NUM_TRAIN_IMAGES = 70000
NUM_VAL_IMAGES = 10000

train_images = sorted(glob(os.path.join(DATA_DIR, "images/training/*")))[:NUM_TRAIN_IMAGES]
train_masks = sorted(glob(os.path.join(DATA_DIR, "annotations/training/*")))[:NUM_TRAIN_IMAGES]
val_images = sorted(glob(os.path.join(DATA_DIR, "images/validation/*")))[
    :NUM_VAL_IMAGES
]
val_masks = sorted(glob(os.path.join(DATA_DIR, "annotations/validation/*")))[
    :NUM_VAL_IMAGES
]

print(len(train_images))
print(len(train_masks))
print(len(val_images))

def read_image(image_path, mask=False):
    image = tf_io.read_file(image_path)
    if mask:
        image = tf_image.decode_png(image, channels=1)
        image.set_shape([None, None, 1])
        image = tf_image.resize(images=image, size=[IMAGE_HEIGHT, IMAGE_WIDTH])
    else:
        image = tf_image.decode_jpeg(image, channels=3)
        image.set_shape([None, None, 3])
        image = tf_image.resize(images=image, size=[IMAGE_HEIGHT, IMAGE_WIDTH])
    return image


def load_data(image_list, mask_list):
    image = read_image(image_list)
    mask = read_image(mask_list, mask=True)
    return image, mask


def data_generator(image_list, mask_list):
    dataset = tf_data.Dataset.from_tensor_slices((image_list, mask_list))
    dataset = dataset.map(load_data, num_parallel_calls=tf_data.AUTOTUNE)
    dataset = dataset.batch(BATCH_SIZE, drop_remainder=True)
    return dataset


train_dataset = data_generator(train_images, train_masks)
val_dataset = data_generator(val_images, val_masks)

model = keras_cv.models.DeepLabV3Plus.from_preset(
    "mobilenet_v3_small_imagenet", num_classes=NUM_CLASSES,load_weights=True
)



model.compile(
    optimizer=keras.optimizers.Adam(learning_rate=0.0001),
    loss=keras.losses.SparseCategoricalCrossentropy(from_logits=False),
    metrics=[keras.metrics.MeanIoU(num_classes=3),keras.metrics.SparseCategoricalAccuracy()]
)
history = model.fit(train_dataset, validation_data=val_dataset, epochs=EPOCHS,verbose=1)

ERROR thrown:

RuntimeError: stack expects each tensor to be equal size, but got [5898240] at entry 0 and [17694720] at entry 1

[sachinprasadhs]

Hi,

Thanks for reporting the issue. To narrow down the issue, could you please check if you are observing same behavior in TensorFlow backend as well?

[github-actions[bot]]

This issue is stale because it has been open for 14 days with no activity. It will be closed if no further activity occurs. Thank you.

[github-actions[bot]]

This issue was closed because it has been inactive for 28 days. Please reopen if you'd like to work on this further.

[google-ml-butler[bot]]

Are you satisfied with the resolution of your issue? Yes No