SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2303.08010 · ICCV · 2023

Window-Based Early-Exit Cascades for Uncertainty Estimation: When Deep Ensembles are More Efficient than Single Models

Christos-Savvas Bouganis, Guoxuan Xia

arXiv · PDF · Open in the Atlas

Code that ran

23 of 36 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 21 have no contract verdict; 1 failure of our instrument; 13 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2303.08010.json · badge: /api/ran/2303.08010.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
keras-team/keras canonical (classifier label)
classifier basis not recorded on the link
22 ran · 13 no recorded run
Guoxoug/window-early-exit community
Papers with Code archive
1 ran · 0 no recorded run
FunctionStatusWhere it lives
ConvNeXtBlock Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/convnext.py
code served (permissive licence) · get_code("9faec97eab071c9e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned function. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
os.environ['KERAS_BACKEND'] = 'tensorflow'

import tensorflow as tf
import numpy as np

# Set seeds for reproducibility
tf.random.set_seed(0)
np.random.seed(0)

# Define ARGS dict with all parameters for ConvNeXtBlock
ARGS = {
    'projection_dim': 64,
    'drop_path_rate': 0.1,
    'layer_scale_init_value': 1e-6,
    'name': 'test_convnext_block'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Head Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/convnext.py
code served (permissive licence) · get_code("8566faf0283e84b5")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned function. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import keras
from keras.src import backend
from keras.src import layers
import numpy as np

ARGS = {
    'num_classes': 10,
    'classifier_activation': 'softmax',
    'name': 'test_head'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
InceptionResNetV2 Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/inception_resnet_v2.py
code served (permissive licence) · get_code("bfe7ed6cde8213a2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Functional. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import keras
from keras.src import backend
from keras.src import layers
from keras.src.layers.layer import Layer
import numpy as np

# Set backend to channels_last (default for most systems)
if backend.image_data_format() != 'channels_last':
    backend.set_image_data_format('channels_last')

ARGS = {
    'include_top': True,
    'weights': None,
    'input_tensor': None,
    'input_shape': (299, 299, 3),
    'pooling': None,
    'classes': 1000,
    'classifier_activation': 'softmax',
    'name': 'inception_resnet_v2'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Functional — a Functional output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Functional output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
InceptionV3 Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/inception_v3.py
code served (permissive licence) · get_code("3562a18df7063eac")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Functional. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import keras
from keras.src import layers
from keras.src import backend

# Set up minimal Keras configuration
backend.set_image_data_format('channels_last')

ARGS = {
    'include_top': True,
    'weights': None,
    'input_tensor': None,
    'input_shape': (299, 299, 3),
    'pooling': None,
    'classes': 1000,
    'classifier_activation': 'softmax',
    'name': 'inception_v3'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Functional — a Functional output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Functional output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
MBConvBlock Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/efficientnet_v2.py
code served (permissive licence) · get_code("6bd8585a9f22e0d2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned function. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import keras
from keras.src import backend
from keras.src import layers
import numpy as np

# Set up a small input tensor for testing
# batch_size=2, height=4, width=4, channels=3
input_tensor = keras.Input(shape=(4, 4, 3))

ARGS = {
    'input_filters': 3,
    'output_filters': 16,
    'expand_ratio': 2,
    'kernel_size': 3,
    'strides': 1,
    'se_ratio': 0.25,
    'bn_momentum': 0.9,
    'activation': 'swish',
    'survival_probability': 0.8,
    'name': 'mbconv_block_'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
MobileNetV2 Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/mobilenet_v2.py
code served (permissive licence) · get_code("3c500f029705ef98")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Functional. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
os.environ['KERAS_BACKEND'] = 'tensorflow'

ARGS = {
    'input_shape': (224, 224, 3),
    'alpha': 1.0,
    'include_top': True,
    'weights': None,
    'input_tensor': None,
    'pooling': None,
    'classes': 1000,
    'classifier_activation': 'softmax',
    'name': 'mobilenetv2'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Functional — a Functional output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Functional output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
block Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/efficientnet.py
code served (permissive licence) · get_code("c84bba05a0e9fe98")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned KerasTensor. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import keras
from keras.src import backend
from keras.src import layers
import numpy as np

# Set up a minimal Keras model input tensor
# The function expects a tensor input with shape (batch, height, width, channels)
# Using batch=2, height=4, width=3, channels=32 for input
inputs = keras.Input(shape=(4, 3, 32), batch_size=2)

ARGS = {
    'inputs': inputs,
    'activation': 'swish',
    'drop_rate': 0.0,
    'name': 'block_',
    'filters_in': 32,
    'filters_out': 16,
    'kernel_size': 3,
    'strides': 1,
    'expand_ratio': 1,
    'se_ratio': 0.0,
    'id_skip': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type KerasTensor — a KerasTensor output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
conv2d_bn Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/inception_resnet_v2.py
code served (permissive licence) · get_code("8c4c044b0da3d59d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned KerasTensor. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import keras
from keras.src import backend
from keras.src import layers
import numpy as np

# Set up a minimal Keras input tensor for the function
# The function expects x to be a tensor that can be passed through Conv2D
input_shape = (2, 4, 4, 3)  # batch=2, height=4, width=4, channels=3
x = keras.Input(shape=(4, 4, 3))

ARGS = {
    'x': x,
    'filters': 32,
    'kernel_size': 3,
    'strides': 1,
    'padding': 'same',
    'activation': 'relu',
    'use_bias': False,
    'name': 'conv2d_bn_test'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type KerasTensor — a KerasTensor output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
conv2d_bn Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/inception_v3.py
code served (permissive licence) · get_code("b55586ffc10b9cf8")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned KerasTensor. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
from tensorflow import keras
from keras.src import backend
from keras.src import layers

# Create a minimal input tensor
# batch=2, channels=3, height=4, width=4
input_tensor = keras.Input(shape=(4, 4, 3))

# Define parameters for conv2d_bn function
filters = 32
num_row = 3
num_col = 3
padding = 'same'
strides = (1, 1)
name = 'test_conv'

ARGS = {
    'x': input_tensor,
    'filters': filters,
    'num_row': num_row,
    'num_col': num_col,
    'padding': padding,
    'strides': strides,
    'name': name
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type KerasTensor — a KerasTensor output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
conv_block Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/densenet.py
code served (permissive licence) · get_code("0dd0e7197a272037")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned EagerTensor, ndim 4, dtype family float, shape (2, 4, 3, 7) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
os.environ['KERAS_BACKEND'] = 'tensorflow'

import tensorflow as tf
import numpy as np
from keras.src import backend
from keras.src import layers

# Set seed for reproducibility
tf.random.set_seed(0)
np.random.seed(0)

# Create a small input tensor
# Shape: (batch=2, height=4, width=3, channels=5)
# Using channels_last format (default in TensorFlow)
x = tf.random.normal((2, 4, 3, 5), dtype=tf.float32)

# Define growth_rate
growth_rate = 2

# Define name
name = "conv_block_0"

ARGS = {
    'x': x,
    'growth_rate': growth_rate,
    'name': name
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 4, dtype family float, shape (2, 4, 3, 7) — shape (2, 4, 3, 7) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
dense_block Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/densenet.py
code served (permissive licence) · get_code("80a7dd835e5af022")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned KerasTensor. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
os.environ['KERAS_BACKEND'] = 'tensorflow'

import tensorflow as tf
from keras.src import backend

# Set up a small input tensor
# batch_size=2, height=4, width=5, channels=3 (channels_last format)
x = tf.keras.Input(shape=(4, 5, 3), batch_size=2)

# Parameters for dense_block
blocks = 2
name = 'dense_block'

ARGS = {
    'x': x,
    'blocks': blocks,
    'name': name,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type KerasTensor — a KerasTensor output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
inception_resnet_block Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/inception_resnet_v2.py
code served (permissive licence) · get_code("072dde7fd28b54c1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned EagerTensor, ndim 4, dtype family float, shape (2, 4, 3, 64) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
os.environ['KERAS_BACKEND'] = 'tensorflow'

import keras
from keras.src import backend
from keras.src import layers
from keras.src.layers.layer import Layer
import tensorflow as tf

# Create a small input tensor (batch_size=2, height=4, width=3, channels=64)
# We use channels=64 because the function expects x.shape[channel_axis] to work
# and the block_type branches produce specific channel counts that get upsampled
input_tensor = tf.keras.Input(shape=(4, 3, 64))

# Create a simple functional model to get a real tensor output
x = input_tensor
x = layers.Conv2D(64, 1, padding='same')(x)
model = tf.keras.Model(inputs=input_tensor, outputs=x)

# Generate a batch of dummy data
dummy_input = tf.random.normal((2, 4, 3, 64))
x_tensor = model(dummy_input)

ARGS = {
    'x': x_tensor,
    'scale': 0.1,
    'block_type': 'block35',
    'block_idx': 0,
    'activation': 'relu'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 4, dtype family float, shape (2, 4, 3, 64) — shape (2, 4, 3, 64) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
leaky_relu Ran
ran on a synthesized input
keras-team/keras/keras/src/activations/activations.py
code served (permissive licence) · get_code("a86f237c3dfe72fb")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned EagerTensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Create a small input tensor
# Using shape (2, 4) for batch size 2, dimension 4
x = torch.randn(2, 4)

# Set negative_slope parameter
negative_slope = 0.2

ARGS = {
    'x': x,
    'negative_slope': negative_slope
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
obtain_input_shape Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/imagenet_utils.py
code served (permissive licence) · get_code("47c0511b0bcb81fb")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family int; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import warnings

# Test case 1: Basic valid input with channels_last format
ARGS = {
    'input_shape': (224, 224, 3),
    'default_size': 224,
    'min_size': 32,
    'data_format': 'channels_last',
    'require_flatten': False,
    'weights': None,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family int — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
preprocess_input Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/imagenet_utils.py
code served (permissive licence) · get_code("fe1fee1f08d6f46d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 4, dtype family float, shape (2, 224, 224, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

ARGS = {
    'x': np.random.randint(0, 256, size=(2, 224, 224, 3), dtype=np.uint8),
    'data_format': 'channels_last',
    'mode': 'caffe'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 4, dtype family float, shape (2, 224, 224, 3) — shape (2, 224, 224, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
preprocess_input Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/inception_v3.py
code served (permissive licence) · get_code("86361b3f206f4002")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 4, dtype family float, shape (2, 4, 4, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# The function takes x (input data) and data_format (optional)
# It calls imagenet_utils.preprocess_input with mode='tf'
# The input x should be image data - typically a numpy array or tensor

# Create a small dummy image batch
# Standard image shape is (batch, height, width, channels) for channels_last
# or (batch, channels, height, width) for channels_first
x = np.random.randint(0, 256, size=(2, 4, 4, 3), dtype=np.uint8)

ARGS = {
    'x': x,
    'data_format': None,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 4, dtype family float, shape (2, 4, 4, 3) — shape (2, 4, 4, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
preprocess_input Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/mobilenet.py
code served (permissive licence) · get_code("b3bae248d68eaa66")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 4, dtype family float, shape (2, 4, 4, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# The function takes an input x (image data) and optional data_format parameter
# x should be a numpy array or similar array-like structure representing image(s)
# data_format can be None, 'channels_first', or 'channels_last'

# Create a small dummy image batch
# Standard image shape: (batch_size, height, width, channels) for channels_last
# or (batch_size, channels, height, width) for channels_first
x = np.random.randint(0, 256, size=(2, 4, 4, 3), dtype=np.uint8)

ARGS = {
    'x': x,
    'data_format': None
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 4, dtype family float, shape (2, 4, 4, 3) — shape (2, 4, 4, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
preprocess_input Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/mobilenet_v2.py
code served (permissive licence) · get_code("dc94020311204486")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 4, dtype family float, shape (2, 224, 224, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

ARGS = {
    'x': np.random.rand(2, 224, 224, 3).astype(np.float32),
    'data_format': None,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 4, dtype family float, shape (2, 224, 224, 3) — shape (2, 224, 224, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
relu6 Ran
ran on a synthesized input
keras-team/keras/keras/src/activations/activations.py
code served (permissive licence) · get_code("2d9c9582df92bc5e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned EagerTensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Create a small input tensor for relu6
# relu6 takes an input tensor and applies relu6 activation
x = torch.randn(2, 4)

ARGS = {
    'x': x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
round_filters Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/efficientnet_v2.py
code served (permissive licence) · get_code("24a8d4e26aa2e281")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

# Define ARGS with all parameters for round_filters function
ARGS = {
    "filters": 32,
    "width_coefficient": 1.0,
    "min_depth": 8,
    "depth_divisor": 8
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
round_repeats Ran
ran, honoured the contract we drafted
keras-team/keras/keras/src/applications/efficientnet_v2.py
code served (permissive licence) · get_code("6cc796a8212b5586")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import math

def round_repeats(repeats, depth_coefficient):
    """Round number of repeats based on depth multiplier."""
    return int(math.ceil(depth_coefficient * repeats))

ARGS = {
    "repeats": 4,
    "depth_coefficient": 1.2
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
transition_block Ran
ran on a synthesized input
keras-team/keras/keras/src/applications/densenet.py
code served (permissive licence) · get_code("ce877225caec4bfe")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned EagerTensor, ndim 4, dtype family float, shape (2, 2, 1, 2) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
os.environ['KERAS_BACKEND'] = 'tensorflow'

import tensorflow as tf
from keras.src import backend, layers

# Set random seed for reproducibility
tf.random.set_seed(0)

# Create a small input tensor
# Shape: (batch=2, height=4, width=3, channels=5)
x = tf.random.normal((2, 4, 3, 5))

# Parameters for the function
reduction = 0.5
name = "transition_block_test"

ARGS = {
    "x": x,
    "reduction": reduction,
    "name": name
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
keras, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 4, dtype family float, shape (2, 2, 1, 2) — shape (2, 2, 1, 2) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
window_threshold_strategy Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Guoxoug/window-early-exit/utils/cascade_utils.py
code served (permissive licence) · get_code("333af359ab2bc74e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create small tensors with batch size 2 and feature dimension 4
data1 = torch.randn(2, 4)
data2 = torch.randn(2, 4)

# Create metrics dict with confidence scores
metrics = {
    'confidence': torch.tensor([0.3, 0.7])
}

# Define all arguments
ARGS = {
    'metrics': metrics,
    'data1': data1,
    'data2': data2,
    'window': [0, 0.5],
    'unc_flip': False,
    'metric_name': 'confidence',
    'tau': None
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
EfficientNet Not yet run
unverified (harvested, no recorded run)
keras-team/keras/keras/src/applications/efficientnet.py
code served (permissive licence) · get_code("b7f04958b9956ab3")
How it ran

no invocation recorded.

EfficientNetB0 Not yet run
unverified (harvested, no recorded run)
keras-team/keras/keras/src/applications/efficientnet.py
code served (permissive licence) · get_code("92506576e1211977")
How it ran

no invocation recorded.

MobileNet Not yet run
unverified (harvested, no recorded run)
keras-team/keras/keras/src/applications/mobilenet.py
code served (permissive licence) · get_code("621b43910a11301f")
How it ran

no invocation recorded.

MobileNetV3 Not yet run
unverified (harvested, no recorded run)
keras-team/keras/keras/src/applications/mobilenet_v3.py
code served (permissive licence) · get_code("2a31220d9622da08")
How it ran

no invocation recorded.

MobileNetV3Large Not yet run
unverified (harvested, no recorded run)
keras-team/keras/keras/src/applications/mobilenet_v3.py
code served (permissive licence) · get_code("418f46abf7d3fddd")
How it ran

no invocation recorded.

MobileNetV3Small Not yet run
unverified (harvested, no recorded run)
keras-team/keras/keras/src/applications/mobilenet_v3.py
code served (permissive licence) · get_code("775b31b0717143c0")
How it ran

no invocation recorded.

PreStem Not yet run
unverified (harvested, no recorded run)
keras-team/keras/keras/src/applications/convnext.py
code served (permissive licence) · get_code("8c030c1dd2cba50c")
How it ran

no invocation recorded.

decode_predictions Not yet run
unverified (harvested, no recorded run)
keras-team/keras/keras/src/applications/imagenet_utils.py
code served (permissive licence) · get_code("d5da2aff6d992a41")
How it ran

no invocation recorded.

decode_predictions Not yet run
unverified (harvested, no recorded run)
keras-team/keras/keras/src/applications/mobilenet.py
code served (permissive licence) · get_code("b8a4e52b0b228ca8")
How it ran

no invocation recorded.

decode_predictions Not yet run
unverified (harvested, no recorded run)
keras-team/keras/keras/src/applications/mobilenet_v2.py
code served (permissive licence) · get_code("fcfce33cd8b78e31")
How it ran

no invocation recorded.

get_name_from_symbol Not yet run
unverified (harvested, no recorded run)
keras-team/keras/keras/src/api_export.py
code served (permissive licence) · get_code("ff36a3530575a45b")
How it ran

no invocation recorded.

get_symbol_from_name Not yet run
unverified (harvested, no recorded run)
keras-team/keras/keras/src/api_export.py
code served (permissive licence) · get_code("b1f9e5129fa630a1")
How it ran

no invocation recorded.

relu Not yet run
unverified (harvested, no recorded run)
keras-team/keras/keras/src/activations/activations.py
code served (permissive licence) · get_code("59c7dca60305d651")
How it ran

no invocation recorded.

Also present in other papers' code

Of 36 samples lifted from this paper's linked repositories, 36 carry a function identity and 0 do not. Of the 36 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 0 later) and 35 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
round_repeats 4 other papers
4 dated earlier
1905.11946 (2019-05)
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
1911.09070 (2019-11)
EfficientDet: Scalable and Efficient Object Detection
1911.12448 (2019-11)
Soft Anchor-Point Object Detection
and 1 more not named here

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Deep Ensembles are a simple, reliable, and effective method of improving both the predictive performance and uncertainty estimates of deep learning approaches. However, they are widely criticised as being computationally expensive, due to the need to deploy multiple independent models. Recent work has challenged this view, showing that for predictive accuracy, ensembles can be more computationally efficient (at inference) than scaling single models within an architecture family. This is achieved by cascading ensemble members via an early-exit approach. In this work, we investigate extending these efficiency gains to tasks related to uncertainty estimation. As many such tasks, e.g. selective classification, are binary classification, our key novel insight is to only pass samples within a window close to the binary decision boundary to later cascade stages. Experiments on ImageNet-scale data across a number of network architectures and uncertainty tasks show that the proposed window-based early-exit approach is able to achieve a superior uncertainty-computation trade-off compared to scaling single models. For example, a cascaded EfficientNet-B2 ensemble is able to achieve similar coverage at 5% risk as a single EfficientNet-B4 with <30% the number of MACs. We also find that cascades/ensembles give more reliable improvements on OOD data vs scaling models up. Code for this work is available at: https://github.com/Guoxoug/window-early-exit.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2303.08010")
get_code_for_paper("2303.08010")
have("2303.08010")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2303.08010.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2303.08010.svg)](https://syntology.ai/paper/2303.08010)

Connect an agent — have() is free.