SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1704.04861 · 2017

MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

Bo Chen, Tobias Weyand, Marco Andreetto, Andrew Howard, Menglong Zhu, Dmitry Kalenichenko, Weijun Wang

arXiv · PDF · Open in the Atlas

Code that ran

53 of 83 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (28 of them constructed an object rather than computing a result; 4 honoured the contract we drafted, 0 violated it and 40 have no contract verdict; 9 failures of our instrument; 30 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/1704.04861.json · badge: /api/ran/1704.04861.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
allenai/dnw community
not in the paper's declared repositories
6 ran · 8 no recorded run
diasirish/mobilenetv3 community
not in the paper's declared repositories
6 ran · 1 no recorded run
Sakib1263/MobileNet-1D-2D-Tensorflow-Keras community
Papers with Code archive
4 ran · 1 no recorded run
ridvansalihkuzu/hyperview_eagleeyes community
not in the paper's declared repositories
3 ran · 2 no recorded run
yakhyo/retinaface-pytorch community
not in the paper's declared repositories
3 ran · 0 no recorded run
deep-learning-algorithm/LightWeightCNN community
not in the paper's declared repositories
3 ran · 0 no recorded run
dronefreak/human-action-classification community
not in the paper's declared repositories
2 ran · 1 no recorded run
Deci-AI/super-gradients community
Papers with Code archive
2 ran · 1 no recorded run
wangvation/torch-mobilenet community
not in the paper's declared repositories
2 ran · 1 no recorded run
aryanasadianuoit/mobilenet_v1 community
not in the paper's declared repositories
2 ran · 0 no recorded run
luzerochen/MobileNet_pytorch community
not in the paper's declared repositories
2 ran · 0 no recorded run
akrapukhin/MobileNetV3 community
not in the paper's declared repositories
2 ran · 0 no recorded run
lyk19940625/MyRFBNet community
not in the paper's declared repositories
2 ran · 0 no recorded run
KaFaiFai/MobileNet-PyTorch-Converter community
not in the paper's declared repositories
2 ran · 0 no recorded run
tensorflow/models community
not in the paper's declared repositories
1 ran · 2 no recorded run
christophmeyer/longboard-pothole-detection community
not in the paper's declared repositories
1 ran · 1 no recorded run
saghiralfasly/VFL-Vehicle-Re-Id community
not in the paper's declared repositories
1 ran · 1 no recorded run
MrRen-sdhm/Embedded_Multi_Object_Detection_CNN community
Papers with Code archive
1 ran · 1 no recorded run
ttruty/facial-feature-mouse-control community
not in the paper's declared repositories
1 ran · 0 no recorded run
marload/ConvNets-TensorFlow2 community
not in the paper's declared repositories
1 ran · 0 no recorded run
jmjeon94/MobileNet-Pytorch community
not in the paper's declared repositories
1 ran · 0 no recorded run
Mayurji/Image-Classification-PyTorch community
Papers with Code archive
1 ran · 0 no recorded run
PolinaDruzhinina/modelnet community
Papers with Code archive
1 ran · 0 no recorded run
idealo/imagededup community
not in the paper's declared repositories
1 ran · 0 no recorded run
Nguyendat-bit/MobilenetV1 community
Papers with Code archive
1 ran · 0 no recorded run
FunctionStatusWhere it lives
ActionClassifier Ran
ran on a synthesized input · constructed an object; no result was computed
dronefreak/human-action-classification/src/hac/image/models/classifier.py
code served (permissive licence) · get_code("de82865faffab4ba")
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 an instance of its own class, ActionClassifier: the run constructed the object and did not call it, so no result was computed. 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 torch
import torch.nn as nn
import timm

torch.manual_seed(0)

ARGS = {
    'model_name': 'mobilenetv3_small_100',
    'num_classes': 40,
    'pretrained': False,
    'freeze_backbone': False,
    'input_type': 'image',
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
timm, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ActionClassifier — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ActionClassifier 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
ChooseTopEdges Ran
ran on a synthesized input · constructed an object; no result was computed
allenai/dnw/models/graphs/mobilenetv1like.py
pointer only (licence: NOASSERTION) · get_code("b57b5ee03a18ef4a")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, allenai/dnw (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Complete Ran
ran on a synthesized input · constructed an object; no result was computed
allenai/dnw/models/graphs/mobilenetv1like.py
pointer only (licence: NOASSERTION) · get_code("08ea70b6df206215")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, allenai/dnw (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Conv Ran
ran on a synthesized input · constructed an object; no result was computed
Deci-AI/super-gradients/src/super_gradients/modules/conv_bn_act_block.py
code served (permissive licence) · get_code("a3a2170c2a66e9f7")
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 an instance of its own class, Conv: the run constructed the object and did not call it, so no result was computed. 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 torch
import torch.nn as nn

torch.manual_seed(0)

# Define a simple activation function for testing
activation = nn.ReLU

# Define the arguments for Conv
ARGS = {
    'input_channels': 3,
    'output_channels': 16,
    'kernel': 3,
    'stride': 1,
    'activation_type': activation,
    'padding': None,
    'groups': None,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
super_gradients, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Conv — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Conv 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
Conv2dNormActivation Ran
ran on a synthesized input · constructed an object; no result was computed
yakhyo/retinaface-pytorch/models/backbones/mobilenetv1.py
code served (permissive licence) · get_code("7267d41ba165eb2f")
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 an instance of its own class, Conv2dNormActivation: the run constructed the object and did not call it, so no result was computed. 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 torch
from torch import nn
from typing import Optional, Callable

torch.manual_seed(0)

ARGS = {
    'in_channels': 3,
    'out_channels': 16,
    'kernel_size': 3,
    'stride': 1,
    'padding': None,
    'groups': 1,
    'norm_layer': torch.nn.BatchNorm2d,
    'activation_layer': torch.nn.LeakyReLU,
    'dilation': 1,
    'inplace': True,
    'negative_slope': 0.1,
    'bias': False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Conv2dNormActivation — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Conv2dNormActivation output the recorded ndim 1 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
ConvBNReLU Ran
ran on a synthesized input · constructed an object; no result was computed
deep-learning-algorithm/LightWeightCNN/py/lib/models/mobilenet/mobilenet.py
code served (permissive licence) · get_code("a9b5f1e8c01481db")
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 an instance of its own class, ConvBNReLU: the run constructed the object and did not call it, so no result was computed. 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 torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    'in_planes': 3,
    'out_planes': 16,
    'kernel_size': 3,
    'stride': 1,
    'padding': 1,
    'groups': 1,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ConvBNReLU — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ConvBNReLU output the recorded ndim 1 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
Conv_2D_DW Ran
ran; our contract draft was wrong, not the code
Sakib1263/MobileNet-1D-2D-Tensorflow-Keras/Codes/MobileNet_2DCNN.py
code served (permissive licence) · get_code("7412089616b18aa5")
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 tensorflow as tf
import numpy as np

# Create a small input tensor
# Shape: (batch_size, height, width, channels)
batch_size = 2
height = 4
width = 4
input_channels = 3

# Create input tensor
inputs = tf.keras.Input(shape=(height, width, input_channels))

# Define arguments
ARGS = {
    'inputs': inputs,
    'model_width': 32,
    'kernel': 3,
    'strides': 1,
    'alpha': 0.5
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
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_2D_block_2 Ran
ran, honoured the contract we drafted
Sakib1263/MobileNet-1D-2D-Tensorflow-Keras/Codes/MobileNet_2DCNN.py
code served (permissive licence) · get_code("badb2ffd2ac2ce23")
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, 4, 16) 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 tensorflow as tf
import numpy as np

# Create a small input tensor for Conv2D
# Shape: (batch_size, height, width, channels)
batch_size = 2
height = 4
width = 4
input_channels = 3

inputs = tf.random.normal((batch_size, height, width, input_channels))

# Define other parameters
model_width = 16  # Number of output filters for Conv2D
kernel = 3  # Kernel size
strides = 1  # Stride
nl = 'HS'  # Nonlinearity type ('HS' for Hard Swish, 'RE' for ReLU with max_value)

ARGS = {
    'inputs': inputs,
    'model_width': model_width,
    'kernel': kernel,
    'strides': strides,
    'nl': nl
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
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, 4, 16) — shape (2, 4, 4, 16) 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
DNW Ran
ran on a synthesized input · constructed an object; no result was computed
allenai/dnw/models/graphs/mobilenetv1like.py
pointer only (licence: NOASSERTION) · get_code("9a38d695e40526d5")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, allenai/dnw (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

DSConv2d Ran
ran on a synthesized input · constructed an object; no result was computed
ttruty/facial-feature-mouse-control/model.py
code served (permissive licence) · get_code("59b3bd4d404d73f1")
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 an instance of its own class, DSConv2d: the run constructed the object and did not call it, so no result was computed. 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 torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    'in_planes': 3,
    'out_planes': 16,
    'kernels_per_layer': 4,
    'groups': 1,
    'old': 0,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DSConv2d — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a DSConv2d 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
DW_Conv_Block Ran
ran on a synthesized input
aryanasadianuoit/mobilenet_v1/MobileNet_V1.py
pointer only (licence: NONE) · get_code("c3e093bdec744091")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, aryanasadianuoit/mobilenet_v1 when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

DepthSepConv Ran
ran on a synthesized input
wangvation/torch-mobilenet/module/mobilenet.py
pointer only (licence: NOASSERTION) · get_code("c0c50cccd77efda7")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, wangvation/torch-mobilenet (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

DepthWiseSeparableConv2d Ran
ran on a synthesized input · constructed an object; no result was computed
yakhyo/retinaface-pytorch/models/backbones/mobilenetv1.py
code served (permissive licence) · get_code("f4f2ed3cbe2afea4")
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 an instance of its own class, DepthWiseSeparableConv2d: the run constructed the object and did not call it, so no result was computed. 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 torch
from torch import nn

torch.manual_seed(0)

# Define ARGS dictionary with all parameters for DepthWiseSeparableConv2d
ARGS = {
    'in_channels': 3,
    'out_channels': 16,
    'stride': 1,
    'norm_layer': nn.BatchNorm2d
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DepthWiseSeparableConv2d — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a DepthWiseSeparableConv2d output the recorded ndim 2 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
DepthwiseConvBlock Ran
ran on a synthesized input · constructed an object; no result was computed
christophmeyer/longboard-pothole-detection/pothole_model/model/model.py
code served (permissive licence) · get_code("0685ca8c06247a04")
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 an instance of its own class, DepthwiseConvBlock: the run constructed the object and did not call it, so no result was computed. 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 tensorflow as tf
import numpy as np

# Create a small input tensor for testing
# Shape: (batch_size, height, width, channels)
input_tensor = tf.random.normal((2, 4, 4, 3))

# Define ARGS dict with all parameters for DepthwiseConvBlock
ARGS = {
    'pointwise_conv_filters': 32,
    'alpha': 0.5,
    'depth_multiplier': 1,
    'strides': (1, 1)
}

# Alternative test case with stride != (1, 1)
ARGS_WITH_STRIDE = {
    'pointwise_conv_filters': 64,
    'alpha': 1.0,
    'depth_multiplier': 1,
    'strides': (2, 2)
}

# Use the basic version for execution
ARGS = {
    'pointwise_conv_filters': 32,
    'alpha': 0.5,
    'depth_multiplier': 1,
    'strides': (1, 1)
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DepthwiseConvBlock — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a DepthwiseConvBlock 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
DepthwiseSeparableConv Ran
ran on a synthesized input · constructed an object; no result was computed
luzerochen/MobileNet_pytorch/mobilenet.py
pointer only (licence: NONE) · get_code("24da924c158c5a6c")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, luzerochen/MobileNet_pytorch when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

DepthwiseSeparableConv Ran
ran on a synthesized input
deep-learning-algorithm/LightWeightCNN/py/lib/models/mobilenet/mobilenet.py
code served (permissive licence) · get_code("8ca5785e149f1a3b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 4, dtype family float; 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 torch
import torch.nn as nn

class ConvBNReLU(nn.Sequential):

    def __init__(self, in_planes, out_planes, kernel_size=3, stride=1, padding=0, groups=1):
        super(ConvBNReLU, self).__init__(nn.Conv2d(in_planes, out_planes, kernel_size, stride=stride, padding=padding, groups=groups, bias=False), nn.BatchNorm2d(out_planes), nn.ReLU6(inplace=True))

class DepthwiseSeparableConv(nn.Sequential):

    def __init__(self, inp, oup, stride):
        super(DepthwiseSeparableConv, self).__init__()
        self.stride = stride
        assert stride in [1, 2]
        padding = 1 if stride == 2 else 0
        layers = []
        layers.extend([ConvBNReLU(inp, oup, kernel_size=3, stride=stride, padding=padding, groups=inp), ConvBNReLU(oup, oup, kernel_size=1, stride=1)])
        self.conv = nn.Sequential(*layers)

    def forward(self, x):
        return self.conv(x)


INIT_ARGS = {
    'inp': 4,
    'oup': 8,
    'stride': 1
}

ARGS = {
    'x': torch.randn(2, 4, 6, 6)
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 4, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
DualTaskClassifier Ran
ran on a synthesized input · constructed an object; no result was computed
dronefreak/human-action-classification/src/hac/image/models/classifier.py
code served (permissive licence) · get_code("22902eb46fbd71ff")
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 an instance of its own class, DualTaskClassifier: the run constructed the object and did not call it, so no result was computed. 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 torch
import torch.nn as nn
from typing import Optional, Dict, Any

torch.manual_seed(0)

class ActionClassifier(nn.Module):
    """Action classifier using pretrained vision models.

    Supports both pose-based and scene-based classification.
    """

    def __init__(self, model_name: str='mobilenetv3_small_100', num_classes: int=40, pretrained: bool=True, freeze_backbone: bool=False, input_type: str='image'):
        """Initialize action classifier.

        Args:
            model_name: timm model name
            (e.g., "mobilenetv3_small_100", "efficientnet_b0")
            num_classes: Number of action classes
            pretrained: Use ImageNet pretrained weights
            freeze_backbone: Freeze backbone weights during training
            input_type: "image" for scene classification, "keypoints" for pose-based
        """
        super().__init__()
        self.model_name = model_name
        self.num_classes = num_classes
        self.input_type = input_type
        if input_type == 'image':
            # Use a simple conv backbone instead of timm to avoid download issues
            self.backbone = nn.Sequential(
                nn.Conv2d(3, 16, kernel_size=3, padding=1),
                nn.ReLU(),
                nn.AdaptiveAvgPool2d((1, 1)),
                nn.Flatten(),
                nn.Linear(16, num_classes)
            )
        else:
            self.backbone = nn.Sequential(nn.Linear(66, 256), nn.ReLU(), nn.Dropout(0.3), nn.Linear(256, 128), nn.ReLU(), nn.Dropout(0.3), nn.Linear(128, 64), nn.ReLU(), nn.Linear(64, num_classes))

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """Forward pass.

        Args:
            x: Input tensor
               - For image: (B, 3, H, W)
               - For keypoints: (B, 66)

        Returns:
            Logits of shape (B, num_classes)
        """
        return self.backbone(x)

    def get_config(self) -> Dict[str, Any]:
        """Get model configuration."""
        return {'model_name': self.model_name, 'num_classes': self.num_classes, 'input_type': self.input_type}


class DualTaskClassifier(nn.Module):
    """
    Dual-task classifier: pose classification + action recognition.
    Mirrors the original v1 architecture but modernized.
    """

    def __init__(self, pose_classifier: Optional[ActionClassifier]=None, action_classifier: Optional[ActionClassifier]=None):
        """Initialize dual-task classifier.

        Args:
            pose_classifier: Model for pose classification (sitting/standing/lying)
            action_classifier: Model for action classification (40 classes)
        """
        super().__init__()
        if pose_classifier is None:
            self.pose_classifier = ActionClassifier(model_name='mobilenetv3_small_100', num_classes=3, pretrained=True, input_type='image')
        else:
            self.pose_classifier = pose_classifier
        if action_classifier is None:
            self.action_classifier = ActionClassifier(model_name='efficientnet_b0', num_classes=40, pretrained=True, input_type='image')
        else:
            self.action_classifier = action_classifier

    def forward(self, x: torch.Tensor, return_both: bool=True) -> Dict[str, torch.Tensor]:
        """Forward pass through both classifiers.

        Args:
            x: Input image tensor (B, 3, H, W)
            return_both: If True, return both pose and action predictions

        Returns:
            Dictionary with 'pose' and/or 'action' logits
        """
        results = {}
        if return_both:
            results['pose'] = self.pose_classifier(x)
            results['action'] = self.action_classifier(x)
        else:
            results['action'] = self.action_classifier(x)
        return results


# Create pose classifier with 3 classes
pose_clf = ActionClassifier(
    model_name='mobilenetv3_small_100',
    num_classes=3,
    pretrained=False,
    freeze_backbone=False,
    input_type='image'
)

# Create action classifier with 40 classes
action_clf = ActionClassifier(
    model_name='efficientnet_b0',
    num_classes=40,
    pretrained=False,
    freeze_backbone=False,
    input_type='image'
)

ARGS = {
    'pose_classifier': pose_clf,
    'action_classifier': action_clf
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
timm, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DualTaskClassifier — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a DualTaskClassifier 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
FineTune Ran
ran on a synthesized input · constructed an object; no result was computed
allenai/dnw/models/graphs/mobilenetv1like.py
pointer only (licence: NOASSERTION) · get_code("8bce1107ee299549")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, allenai/dnw (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

FullyConnected Ran
ran on a synthesized input · constructed an object; no result was computed
ridvansalihkuzu/hyperview_eagleeyes/experimental_2/backbone_models/mobile_vit.py
code served (permissive licence) · get_code("a7cb0523df999a6c")
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 an instance of its own class, FullyConnected: the run constructed the object and did not call it, so no result was computed. 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 tensorflow as tf
from tensorflow.keras import layers
from tensorflow import keras

ARGS = {
    'hidden_units': [16, 8],
    'dropout_rate': 0.2
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type FullyConnected — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a FullyConnected 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
Graph Ran
ran on a synthesized input · constructed an object; no result was computed
allenai/dnw/models/graphs/mobilenetv1like.py
pointer only (licence: NOASSERTION) · get_code("bd3e76e24b52bee7")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, allenai/dnw (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Hswish Ran
ran on a synthesized input · constructed an object; no result was computed
akrapukhin/MobileNetV3/models.py
pointer only (licence: NONE) · get_code("ef6ceebbb94bba13")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, akrapukhin/MobileNetV3 when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

IRLB Ran
ran on a synthesized input · constructed an object; no result was computed
diasirish/mobilenetv3/model.py
pointer only (licence: NONE) · get_code("f515f64fa287f6a7")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, diasirish/mobilenetv3 when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

IRLB_SE Ran
ran on a synthesized input · constructed an object; no result was computed
diasirish/mobilenetv3/model.py
pointer only (licence: NONE) · get_code("74429d073cb58967")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, diasirish/mobilenetv3 when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

InvertedRes Ran
ran on a synthesized input · constructed an object; no result was computed
ridvansalihkuzu/hyperview_eagleeyes/experimental_2/backbone_models/mobile_vit.py
code served (permissive licence) · get_code("c92923bdf26c593d")
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 an instance of its own class, InvertedRes: the run constructed the object and did not call it, so no result was computed. 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 tensorflow as tf
from tensorflow.keras import layers
from tensorflow import keras
import numpy as np

class InvertedRes(layers.Layer):

    def __init__(self, expand_channels, output_channels, strides=1):
        super().__init__(name='inverted_res')
        self.output_channels = output_channels
        self.strides = strides
        self.expand = keras.Sequential([layers.Conv2D(expand_channels, 1, padding='same', use_bias=False), layers.BatchNormalization(), layers.Activation('swish')])
        self.dw_conv = keras.Sequential([layers.DepthwiseConv2D(3, strides=strides, padding='same', use_bias=False), layers.BatchNormalization(), layers.Activation('swish')])
        self.pw_conv = keras.Sequential([layers.Conv2D(output_channels, 1, padding='same', use_bias=False), layers.BatchNormalization()])

    def call(self, x):
        o = self.expand(x)
        o = self.dw_conv(o)
        o = self.pw_conv(o)
        if self.strides == 1 and o.shape[-1] == self.output_channels:
            return o + x
        return o

ARGS = {
    'expand_channels': 16,
    'output_channels': 8,
    'strides': 1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type InvertedRes — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a InvertedRes 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
MobileNet Ran
ran on a synthesized input · constructed an object; no result was computed
luzerochen/MobileNet_pytorch/mobilenet.py
pointer only (licence: NONE) · get_code("7189568b40eafada")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, luzerochen/MobileNet_pytorch when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

MobileNet Ran
ran on a synthesized input
lyk19940625/MyRFBNet/models/RFB_Net_mobile.py
code served (permissive licence) · get_code("f4e54c410555f41d")
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 list. 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 torch
import torch.nn as nn

torch.manual_seed(0)

# MobileNet() takes no parameters, so ARGS is empty
ARGS = {}
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 list — a list 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
MobileNet Ran
ran on a synthesized input · constructed an object; no result was computed
deep-learning-algorithm/LightWeightCNN/py/lib/models/mobilenet/mobilenet.py
code served (permissive licence) · get_code("2f65608641fc45af")
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 an instance of its own class, MobileNet: the run constructed the object and did not call it, so no result was computed. 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 torch
import torch.nn as nn

torch.manual_seed(0)

# The MobileNet constructor takes only one parameter: num_classes
# We'll create a simple value for it

ARGS = {
    'num_classes': 10
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MobileNet — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a MobileNet 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
MobileNet Ran
ran on a synthesized input
KaFaiFai/MobileNet-PyTorch-Converter/model/mobile_net.py
pointer only (licence: NONE) · get_code("52f5705b74081b33")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, KaFaiFai/MobileNet-PyTorch-Converter when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

MobileNet Ran
ran on a synthesized input
marload/ConvNets-TensorFlow2/models/MobileNet.py
code served (permissive licence) · get_code("3feb913ca0fba9d5")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned EagerTensor, ndim 2, dtype family float; 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 tensorflow as tf
from tensorflow.keras import layers, Sequential, Model
import numpy as np

class MobileNet(Model):

    def __init__(self, num_classes, alpha=1, input_shape=(32, 32, 3)):
        super(MobileNet, self).__init__()
        self.conv1 = Sequential([layers.Input(input_shape), layers.Conv2D(32, (3, 3), strides=2, padding='same', activation='relu'), layers.SeparableConv2D(64, (3, 3), strides=1, padding='same', depth_multiplier=alpha, activation='relu')])
        self.conv2 = Sequential([layers.SeparableConv2D(128, (3, 3), strides=1, padding='same', depth_multiplier=alpha, activation='relu'), layers.SeparableConv2D(128, (3, 3), strides=1, padding='same', depth_multiplier=alpha, activation='relu')])
        self.conv3 = Sequential([layers.SeparableConv2D(1256, (3, 3), strides=1, padding='same', depth_multiplier=alpha, activation='relu'), layers.SeparableConv2D(256, (3, 3), strides=1, padding='same', depth_multiplier=alpha, activation='relu')])
        self.conv4 = Sequential([layers.SeparableConv2D(512, (3, 3), strides=1, padding='same', depth_multiplier=alpha, activation='relu'), layers.SeparableConv2D(512, (3, 3), strides=1, padding='same', depth_multiplier=alpha, activation='relu'), layers.SeparableConv2D(512, (3, 3), strides=1, padding='same', depth_multiplier=alpha, activation='relu'), layers.SeparableConv2D(512, (3, 3), strides=1, padding='same', depth_multiplier=alpha, activation='relu'), layers.SeparableConv2D(512, (3, 3), strides=1, padding='same', depth_multiplier=alpha, activation='relu'), layers.SeparableConv2D(512, (3, 3), strides=1, padding='same', depth_multiplier=alpha, activation='relu')])
        self.conv5 = Sequential([layers.SeparableConv2D(1024, (3, 3), strides=1, padding='same', depth_multiplier=alpha, activation='relu'), layers.SeparableConv2D(1024, (3, 3), strides=1, padding='same', depth_multiplier=alpha, activation='relu')])
        self.ap = layers.AveragePooling2D((7, 7), strides=1)
        self.flat = layers.Flatten()
        self.fc = layers.Dense(num_classes, activation='softmax')

    def call(self, inputs, training=False):
        x = self.conv1(inputs)
        x = self.conv2(x)
        x = self.conv3(x)
        x = self.conv4(x)
        x = self.conv5(x)
        x = self.ap(x)
        x = self.flat(x)
        x = self.fc(x)
        return x

INIT_ARGS = {
    'num_classes': 5,
    'alpha': 1,
    'input_shape': (16, 16, 3)
}

ARGS = {
    'inputs': np.random.randn(2, 16, 16, 3).astype(np.float32)
}

model = MobileNet(**INIT_ARGS)
output = model(**ARGS)
print(f"Output shape: {output.shape}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
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 2, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
MobileNetV1 Ran
ran on a synthesized input · constructed an object; no result was computed
yakhyo/retinaface-pytorch/models/backbones/mobilenetv1.py
code served (permissive licence) · get_code("add0cbc0696d05ab")
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 an instance of its own class, MobileNetV1: the run constructed the object and did not call it, so no result was computed. 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 torch
import torch.nn as nn

torch.manual_seed(0)

# For MobileNetV1, we need to construct a test input tensor
# The function expects:
# - width_mult: float (default 0.25)
# - num_classes: int (default 1000)

ARGS = {
    "width_mult": 0.25,
    "num_classes": 10,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MobileNetV1 — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a MobileNetV1 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
MobileNetV1 Ran
ran on a synthesized input
jmjeon94/MobileNet-Pytorch/MobileNetV1.py
pointer only (licence: NONE) · get_code("a9d9d4d8531f8f6b")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, jmjeon94/MobileNet-Pytorch when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

MobileNetV1 Ran
ran on a synthesized input
Mayurji/Image-Classification-PyTorch/MobileNetV1.py
pointer only (licence: GPL-3.0) · get_code("e2d390adb77544cf")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, Mayurji/Image-Classification-PyTorch, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

MobileNetV1 Ran
ran on a synthesized input
aryanasadianuoit/mobilenet_v1/MobileNet_V1.py
pointer only (licence: NONE) · get_code("615563d99451ac35")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, aryanasadianuoit/mobilenet_v1 when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

MobileNetv1 Ran
ran on a synthesized input · constructed an object; no result was computed
PolinaDruzhinina/modelnet/models/mobilenetv1.py
pointer only (licence: NONE) · get_code("e97ce510c599fad7")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, PolinaDruzhinina/modelnet when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

MobilenetV3 Ran
ran on a synthesized input
idealo/imagededup/imagededup/utils/models.py
code served (permissive licence) · get_code("dcbcaf78bf35b6e3")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 2, dtype family float; 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 torch
from torchvision import models
from torchvision.transforms import transforms

# Mock weights to avoid downloads
class MockWeights:
    IMAGENET1K_V1 = None

# Mock mobilenet_v3_small to avoid downloads
def mock_mobilenet_v3_small(weights=None):
    class MockFeatures(torch.nn.Module):
        def __init__(self):
            super().__init__()
            self.conv = torch.nn.Conv2d(3, 4, kernel_size=3, stride=2, padding=1)

        def forward(self, x):
            return self.conv(x)

    class MockAvgPool(torch.nn.Module):
        def __init__(self):
            super().__init__()

        def forward(self, x):
            return torch.nn.functional.adaptive_avg_pool2d(x, (1, 1))

    mock = torch.nn.Module()
    mock.features = MockFeatures()
    mock.avgpool = MockAvgPool()
    return mock

# Patch models.mobilenet_v3_small
models.mobilenet_v3_small = mock_mobilenet_v3_small
models.MobileNet_V3_Small_Weights = MockWeights

# INIT_ARGS for MobilenetV3.__init__
INIT_ARGS = {}

# ARGS for MobilenetV3.forward
ARGS = {
    'x': torch.randn(2, 3, 32, 32)  # batch=2, channels=3, height=32, width=32
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch, torchvision — 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 — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Mobilenet_V1 Ran
ran on a synthesized input · constructed an object; no result was computed
Nguyendat-bit/MobilenetV1/model.py
code served (permissive licence) · get_code("fec14b4fe41163a7")
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 an instance of its own class, Mobilenet_V1: the run constructed the object and did not call it, so no result was computed. 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 tensorflow as tf
from tensorflow.keras import models, Sequential, layers
from tensorflow.keras.layers import (
    InputLayer, Conv2D, BatchNormalization, ReLU, DepthwiseConv2D,
    GlobalAveragePooling2D, Dropout, Dense
)

ARGS = {
    'classes': 10,
    'alpha': 0.5,
    'rho': 0.75,
    'droppout': 0.001,
    'img_size': (224, 224),
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Mobilenet_V1 — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Mobilenet_V1 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
Mobilenet_v3_large Ran
ran on a synthesized input · constructed an object; no result was computed
akrapukhin/MobileNetV3/models.py
pointer only (licence: NONE) · get_code("2f387b881c39f3a7")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, akrapukhin/MobileNetV3 when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

SE_layer Ran
ran on a synthesized input
diasirish/mobilenetv3/model.py
pointer only (licence: NONE) · get_code("ef8ee86c3f6eb804")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, diasirish/mobilenetv3 when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

SeparableConv Ran
ran on a synthesized input
KaFaiFai/MobileNet-PyTorch-Converter/model/mobile_net.py
pointer only (licence: NONE) · get_code("a3f785b105c680fd")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, KaFaiFai/MobileNet-PyTorch-Converter when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Transformer Ran
ran on a synthesized input · constructed an object; no result was computed
ridvansalihkuzu/hyperview_eagleeyes/experimental_2/backbone_models/mobile_vit.py
code served (permissive licence) · get_code("3a76f6827f176b38")
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 an instance of its own class, Transformer: the run constructed the object and did not call it, so no result was computed. 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 tensorflow as tf
from tensorflow.keras import layers
from tensorflow import keras

class FullyConnected(layers.Layer):

    def __init__(self, hidden_units, dropout_rate):
        super().__init__(name='fully_connected')
        l = []
        for units in hidden_units:
            l.append(layers.Dense(units, activation=tf.nn.swish))
            l.append(layers.Dropout(dropout_rate))
        self.mlp = keras.Sequential(l)

    def call(self, x):
        return self.mlp(x)

class Transformer(layers.Layer):

    def __init__(self, projection_dim, heads=2):
        super().__init__(name='transformer')
        self.norm1 = layers.LayerNormalization(epsilon=1e-06)
        self.attention = layers.MultiHeadAttention(num_heads=heads, key_dim=projection_dim, dropout=0.1)
        self.norm2 = layers.LayerNormalization(epsilon=1e-06)

    def build(self, input_shape):
        self.mlp = FullyConnected([input_shape[-1] * 2, input_shape[-1]], dropout_rate=0.1)

    def call(self, x):
        x1 = self.norm1(x)
        att = self.attention(x1, x1)
        x2 = x + att
        x3 = self.norm2(x2)
        x3 = self.mlp(x3)
        return x3 + x2

# Create test input tensor
test_input = tf.random.normal((2, 4))  # batch_size=2, feature_dim=4

# Create Transformer instance and build it
transformer = Transformer(projection_dim=4, heads=2)
transformer.build(test_input.shape)

# Define arguments dict for Transformer.__init__
ARGS = {
    'projection_dim': 4,
    'heads': 2
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Transformer — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Transformer 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
_depthwise_conv_block Ran
ran; our contract draft was wrong, not the code
saghiralfasly/VFL-Vehicle-Re-Id/models/Mob_VFL.py
pointer only (licence: NONE) · get_code("89abfc786d1aa21c")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, saghiralfasly/VFL-Vehicle-Re-Id when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

_fixed_padding Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
tensorflow/models/research/object_detection/models/keras_models/mobilenet_v1.py
pointer only (licence: NOASSERTION) · get_code("cdf860d5cf540b7e")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, tensorflow/models (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

_make_divisible Ran
ran, honoured the contract we drafted
diasirish/mobilenetv3/model.py
pointer only (licence: NONE) · get_code("0d8f0dbec2595af8")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, diasirish/mobilenetv3 when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

_make_divisible Ran
ran, honoured the contract we drafted
wangvation/torch-mobilenet/module/mobilenet.py
pointer only (licence: NOASSERTION) · get_code("4369467852623e56")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, wangvation/torch-mobilenet (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

_squeeze Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Sakib1263/MobileNet-1D-2D-Tensorflow-Keras/Codes/MobileNet_2DCNN.py
code served (permissive licence) · get_code("4af315db2e30f663")
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 EagerTensor, ndim 4, dtype family float, shape (2, 4, 3, 5) 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 tensorflow as tf
import numpy as np

# Create a small 4D input tensor (batch_size=2, height=4, width=3, channels=5)
# This represents a typical feature map from a CNN
inputs = tf.constant(
    np.random.randn(2, 4, 3, 5).astype(np.float32)
)

ARGS = {
    'inputs': inputs
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
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, 5) — shape (2, 4, 3, 5) 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
autopad Ran
ran, honoured the contract we drafted
Deci-AI/super-gradients/src/super_gradients/modules/conv_bn_act_block.py
code served (permissive licence) · get_code("3679480e08fecc84")
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

torch.manual_seed(0)

# The autopad function takes:
# - kernel: either an int or a sequence of ints
# - padding: optional, can be None or a value

# Test case 1: kernel as int, padding=None
kernel_int = 3
padding_none = None

ARGS = {
    "kernel": kernel_int,
    "padding": padding_none
}
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
bottleneck_block_2 Ran
ran; our contract draft was wrong, not the code
Sakib1263/MobileNet-1D-2D-Tensorflow-Keras/Codes/MobileNet_2DCNN.py
code served (permissive licence) · get_code("053a423893bef6e3")
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, 4, 16) 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 tensorflow as tf
import numpy as np

# Create a small input tensor (batch_size=2, height=4, width=4, channels=3)
inputs = tf.keras.Input(shape=(4, 4, 3), batch_size=2)

# Build a minimal model to get a tensor output
model = tf.keras.Sequential([
    tf.keras.layers.Conv2D(3, 1, padding='same', input_shape=(4, 4, 3))
])
inputs_tensor = model(np.random.randn(2, 4, 4, 3).astype(np.float32))

ARGS = {
    'inputs': inputs_tensor,
    'filters': 16,
    'kernel': 3,
    'e': 6,
    's': 1,
    'squeeze': True,
    'nl': 'HS',
    'alpha': 1.0
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
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, 4, 16) — shape (2, 4, 4, 16) 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
conv_bn Ran
ran; our contract draft was wrong, not the code
this paper's copy was not recorded; identical code first harvested from d-li14/octconv.pytorch
pointer only · get_code("2f7853ff01cbbc29")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

conv_dw Ran
ran; our contract draft was wrong, not the code
lyk19940625/MyRFBNet/models/RFB_Net_mobile.py
code served (permissive licence) · get_code("91562b7b744b0715")
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 Sequential. 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 torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "inp": 4,
    "oup": 8,
    "stride": 1,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Sequential — a Sequential output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Sequential output the recorded ndim 1 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
create_ssd_anchors Ran
ran; our contract draft was wrong, not the code
MrRen-sdhm/Embedded_Multi_Object_Detection_CNN/training folder for caffe on PC/MobileNetV2-SSDLite/gen_model.py
pointer only (licence: NONE) · get_code("c6746c730a18b75a")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MrRen-sdhm/Embedded_Multi_Object_Detection_CNN when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_conv Ran
ran; our contract draft was wrong, not the code
allenai/dnw/models/graphs/mobilenetv1like.py
pointer only (licence: NOASSERTION) · get_code("104c50ff520e011e")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, allenai/dnw (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

h_sigmoid Ran
ran on a synthesized input · constructed an object; no result was computed
diasirish/mobilenetv3/model.py
pointer only (licence: NONE) · get_code("839f2594c84662ff")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, diasirish/mobilenetv3 when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

h_swish Ran
ran on a synthesized input · constructed an object; no result was computed
diasirish/mobilenetv3/model.py
pointer only (licence: NONE) · get_code("a61239e09f031487")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, diasirish/mobilenetv3 when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Block Not yet run
unverified (harvested, no recorded run)
allenai/dnw/models/graphs/mobilenetv1like.py
pointer only (licence: NOASSERTION) · get_code("67e69022e2fdbfbc")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, allenai/dnw (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

DNWNoUpdate Not yet run
unverified (harvested, no recorded run)
allenai/dnw/models/graphs/mobilenetv1like.py
pointer only (licence: NOASSERTION) · get_code("f78b91459b1e5597")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, allenai/dnw (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Generator Not yet run
unverified (harvested, no recorded run)
MrRen-sdhm/Embedded_Multi_Object_Detection_CNN/training folder for caffe on PC/MobileNetV2-SSDLite/gen_model.py
pointer only (licence: NONE) · get_code("e172ae9024954d44")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, MrRen-sdhm/Embedded_Multi_Object_Detection_CNN when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Linear Not yet run
unverified (harvested, no recorded run)
allenai/dnw/models/graphs/mobilenetv1like.py
pointer only (licence: NOASSERTION) · get_code("31a526c27fe6c762")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, allenai/dnw (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

MobileNet Not yet run
unverified (harvested, no recorded run)
BillyGun27/mobilenet_model/mobilenet.py
code served (permissive licence) · get_code("c09c47e3f1b1343a")
How it ran

no invocation recorded.

MobileNet Not yet run
unverified (harvested, no recorded run)
Sakib1263/MobileNet-1D-2D-Tensorflow-Keras/Codes/MobileNet_2DCNN.py
code served (permissive licence) · get_code("306134e17b7afcd1")
How it ran

no invocation recorded.

MobileNetV1 Not yet run
unverified (harvested, no recorded run)
wangvation/torch-mobilenet/module/mobilenet.py
pointer only (licence: NOASSERTION) · get_code("fdf90307afecd932")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, wangvation/torch-mobilenet (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Showing the first 60. An agent gets every one from get_harvested_code_for_paper("1704.04861").

Also present in other papers' code

Of 83 samples lifted from this paper's linked repositories, 83 carry a function identity and 0 do not. Of the 83 distinct functions with an identity, 6 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 6 later) and 77 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
conv_bn
brick of this paper's assembly: model
16 other papers
16 later
1807.10221 (2018-07)
Unified Perceptual Parsing for Scene Understanding
1904.05049 (2019-04)
Drop an Octave: Reducing Spatial Redundancy in Convolutional Neural Networks with Octave Convolution
1904.12368 (2019-04)
Towards Efficient Model Compression via Learned Global Ranking
and 13 more not named here
_make_divisible
brick of this paper's assembly: helper
11 other papers
11 later · 7 only as an assembly brick
1801.04381 (2018-01)
MobileNetV2: Inverted Residuals and Linear Bottlenecks
1802.02611 (2018-02)
Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
1905.02244 (2019-05)
Searching for MobileNetV3
and 8 more not named here
relu6
brick of this paper's assembly: helper
6 other papers
2 dated earlier, 4 later
1512.03385 (2015-12)
Deep Residual Learning for Image Recognition
1612.03716 (2016-12)
COCO-Stuff: Thing and Stuff Classes in Context
1801.04381 (2018-01)
MobileNetV2: Inverted Residuals and Linear Bottlenecks
and 3 more not named here
Conv2dNormActivation
brick of this paper's assembly: helper
1 other paper
1 later
1801.04381 (2018-01)
MobileNetV2: Inverted Residuals and Linear Bottlenecks
SupportsReplaceInputChannels
brick of this paper's assembly: helper
1 other paper
1 later
2104.13188 (2021-04)
Rethinking BiSeNet For Real-time Semantic Segmentation
_make_divisible
brick of this paper's assembly: helper
1 other paper
1 later
1905.02244 (2019-05)
Searching for MobileNetV3

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

More are linked than shown here.

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

Abstract

We present a class of efficient models called MobileNets for mobile and embedded vision applications. MobileNets are based on a streamlined architecture that uses depthwise separable convolutions to build light weight deep neural networks. We introduce two simple global hyperparameters that efficiently trade off between latency and accuracy. These hyper-parameters allow the model builder to choose the right sized model for their application based on the constraints of the problem. We present extensive experiments on resource and accuracy tradeoffs and show strong performance compared to other popular models on ImageNet classification. We then demonstrate the effectiveness of MobileNets across a wide range of applications and use cases including object detection, finegrain classification, face attributes and large scale geo-localization.

For agents

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

get_harvested_code_for_paper("1704.04861")
get_code_for_paper("1704.04861")
have("1704.04861")

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

curl https://syntology.ai/api/ran/1704.04861.json

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

[![Syntology run record](https://syntology.ai/api/ran/1704.04861.svg)](https://syntology.ai/paper/1704.04861)

Connect an agent — have() is free.