SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1611.08050 · 2016

Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields

arXiv · PDF · Open in the Atlas

Code that ran

18 of 23 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, 1 violated it and 13 have no contract verdict; 3 failures of our instrument; 5 no recorded run), measured 2026-11-21; 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/1611.08050.json · badge: /api/ran/1611.08050.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
ibm/max-human-pose-estimator community (classifier label)
classifier basis not recorded on the link
13 ran · 5 no recorded run
DavHoffmann/LearningToTrainWithSyntheticHumans extension (classifier label)
classifier basis: classifier_2025
2 ran · 0 no recorded run
Xingyu-Romantic/RMPose_PAF reimplementation (classifier label)
classifier basis: classifier_2025
2 ran · 0 no recorded run
MSeeker1340/Vision2018-Pose community (classifier label)
classifier basis not recorded on the link
1 ran · 0 no recorded run
FunctionStatusWhere it lives
arrayCast Ran
ran on a synthesized input
ibm/max-human-pose-estimator/core/tf_pose/slidingwindow/ArrayUtils.py
code served (permissive licence) · get_code("20375b6e697cb883")
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 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 numpy as np
import torch
import math
import mmap
import tempfile
import psutil

torch.manual_seed(0)

# Create a source numpy array with a specific shape and dtype
source_array = np.array([[1.5, 2.3, 3.1, 4.7],
                         [5.2, 6.9, 7.1, 8.4]], dtype=np.float64)

# Define the target dtype for casting
target_dtype = np.float32

ARGS = {
    "source": source_array,
    "dtype": target_dtype
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, psutil, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, 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
arrayFactory Ran
ran on a synthesized input
ibm/max-human-pose-estimator/core/tf_pose/slidingwindow/ArrayUtils.py
code served (permissive licence) · get_code("967290f028bec7f8")
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 3, dtype family float, shape (4, 5, 6) 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
import mmap
import tempfile
import numpy as np
import psutil

def _requiredSize(shape, dtype):
    """
    Determines the number of bytes required to store a NumPy array with
    the specified shape and datatype.
    """
    return math.floor(np.prod(np.asarray(shape, dtype=np.uint64)) * np.dtype(dtype).itemsize)

class TempfileBackedArray(np.ndarray):
    """
    A NumPy ndarray that uses a memory-mapped temp file as its backing
    """

    def __new__(subtype, shape, dtype=float, buffer=None, offset=0, strides=None, order=None, info=None):
        numBytes = _requiredSize(shape, dtype)
        tempFile = tempfile.TemporaryFile()
        tempFile.truncate(numBytes)
        buf = mmap.mmap(tempFile.fileno(), numBytes, access=mmap.ACCESS_WRITE)
        obj = super(TempfileBackedArray, subtype).__new__(subtype, shape, dtype, buf, 0, None, order)
        obj._file = tempFile
        return obj

    def __array_finalize__(self, obj):
        if obj is None:
            return
        self._file = getattr(obj, '_file', None)

def arrayFactory(shape, dtype=float):
    """
    Creates a new ndarray of the specified shape and datatype, storing
    it in memory if there is sufficient available space or else using
    a memory-mapped temporary file to provide the underlying buffer.
    """
    requiredBytes = _requiredSize(shape, dtype)
    vmem = psutil.virtual_memory()
    if vmem.available > requiredBytes:
        return np.ndarray(shape=shape, dtype=dtype)
    else:
        return TempfileBackedArray(shape=shape, dtype=dtype)


ARGS = {
    "shape": (4, 5, 6),
    "dtype": np.float32
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, psutil — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family float, shape (4, 5, 6) — shape (4, 5, 6) 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
batchWindows Ran
ran on a synthesized input
ibm/max-human-pose-estimator/core/tf_pose/slidingwindow/Batching.py
code served (permissive licence) · get_code("92fd1e2b70b24c05")
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 list, ndim 3, 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 numpy as np

# Create sample windows data
# windows should be a list or array-like structure that can be converted to numpy array
windows = [
    np.array([1, 2, 3, 4]),
    np.array([5, 6, 7, 8]),
    np.array([9, 10, 11, 12]),
    np.array([13, 14, 15, 16]),
    np.array([17, 18, 19, 20]),
    np.array([21, 22, 23, 24]),
]

# batchSize should be an integer that divides evenly into len(windows)
batchSize = 2

ARGS = {
    "windows": windows,
    "batchSize": batchSize,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 3, dtype family int — a list 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
compute_gt_teacher Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
DavHoffmann/LearningToTrainWithSyntheticHumans/tf_pose/train_with_adversarial_teacher.py
pointer only (licence: NOASSERTION) · get_code("f4d54c84f9d6d11f")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, DavHoffmann/LearningToTrainWithSyntheticHumans (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.

cropRect Ran
ran on a synthesized input
ibm/max-human-pose-estimator/core/tf_pose/slidingwindow/RectangleUtils.py
code served (permissive licence) · get_code("eb150ddc7184f446")
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 torch

torch.manual_seed(0)

# rect is a tuple of (x, y, w, h)
rect = (10, 20, 100, 50)

# cropTop, cropBottom, cropLeft, cropRight are pixel values
cropTop = 5
cropBottom = 3
cropLeft = 2
cropRight = 4

ARGS = {
    "rect": rect,
    "cropTop": cropTop,
    "cropBottom": cropBottom,
    "cropLeft": cropLeft,
    "cropRight": cropRight,
}
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 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
cropRectEqually Ran
ran on a synthesized input
ibm/max-human-pose-estimator/core/tf_pose/slidingwindow/RectangleUtils.py
code served (permissive licence) · get_code("630ac0a9f50488c4")
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 torch

# Define the arguments for cropRectEqually
rect = (10, 20, 100, 80)  # (x, y, w, h) - a rectangle at (10,20) with width 100 and height 80
cropping = 5  # crop 5 pixels from all sides

ARGS = {
    "rect": rect,
    "cropping": cropping
}
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 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
generate Ran
ran on a synthesized input
ibm/max-human-pose-estimator/core/tf_pose/slidingwindow/SlidingWindow.py
code served (permissive licence) · get_code("cdb1c9715688f9a1")
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 list, ndim 1, 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 numpy as np
import math

torch.manual_seed(0)

class DimOrder(object):
    """
    Represents the order of the dimensions in a dataset's shape.
    """
    ChannelHeightWidth = ['c', 'h', 'w']
    HeightWidthChannel = ['h', 'w', 'c']

class SlidingWindow(object):
    """
    Represents a single window into a larger dataset.
    """

    def __init__(self, x, y, w, h, dimOrder, transform=None):
        """
        Creates a new window with the specified dimensions and transform
        """
        self.x = x
        self.y = y
        self.w = w
        self.h = h
        self.dimOrder = dimOrder
        self.transform = transform

    def apply(self, matrix):
        """
        Slices the supplied matrix and applies any transform bound to this window
        """
        view = matrix[self.indices()]
        return self.transform(view) if self.transform is not None else view

    def getRect(self):
        """
        Returns the window bounds as a tuple of (x,y,w,h)
        """
        return (self.x, self.y, self.w, self.h)

    def setRect(self, rect):
        """
        Sets the window bounds from a tuple of (x,y,w,h)
        """
        self.x, self.y, self.w, self.h = rect

    def indices(self, includeChannel=True):
        """
        Retrieves the indices for this window as a tuple of slices
        """
        if self.dimOrder == DimOrder.HeightWidthChannel:
            return (slice(self.y, self.y + self.h), slice(self.x, self.x + self.w))
        elif self.dimOrder == DimOrder.ChannelHeightWidth:
            if includeChannel is True:
                return (slice(None, None), slice(self.y, self.y + self.h), slice(self.x, self.x + self.w))
            else:
                return (slice(self.y, self.y + self.h), slice(self.x, self.x + self.w))
        else:
            raise ValueError('Unsupported order of dimensions: ' + str(self.dimOrder))

    def __str__(self):
        return '(' + str(self.x) + ',' + str(self.y) + ',' + str(self.w) + ',' + str(self.h) + ')'

    def __repr__(self):
        return self.__str__()

def generateForSize(width, height, dimOrder, maxWindowSizeW, maxWindowSizeH, overlapPercent, transforms=[]):
    """
    Generates a set of sliding windows for a dataset with the specified dimensions and order.
    """
    windowSizeX = min(maxWindowSizeW, width)
    windowSizeY = min(maxWindowSizeH, height)
    windowOverlapX = int(math.floor(windowSizeX * overlapPercent))
    windowOverlapY = int(math.floor(windowSizeY * overlapPercent))
    stepSizeX = windowSizeX - windowOverlapX
    stepSizeY = windowSizeY - windowOverlapY
    lastX = width - windowSizeX
    lastY = height - windowSizeY
    xOffsets = list(range(0, lastX + 1, stepSizeX))
    yOffsets = list(range(0, lastY + 1, stepSizeY))
    if len(xOffsets) == 0 or xOffsets[-1] != lastX:
        xOffsets.append(lastX)
    if len(yOffsets) == 0 or yOffsets[-1] != lastY:
        yOffsets.append(lastY)
    windows = []
    for xOffset in xOffsets:
        for yOffset in yOffsets:
            for transform in [None] + transforms:
                windows.append(SlidingWindow(x=xOffset, y=yOffset, w=windowSizeX, h=windowSizeY, dimOrder=dimOrder, transform=transform))
    return windows

def generate(data, dimOrder, maxWindowSizeW, maxWindowSizeH, overlapPercent, transforms=[]):
    """
    Generates a set of sliding windows for the specified dataset.
    """
    width = data.shape[dimOrder.index('w')]
    height = data.shape[dimOrder.index('h')]
    return generateForSize(width, height, dimOrder, maxWindowSizeW, maxWindowSizeH, overlapPercent, transforms)


# Create a simple transform function
def simple_transform(x):
    return x * 2

# Create test data with ChannelHeightWidth ordering (c, h, w)
# Shape: (channels=3, height=8, width=16)
data = torch.randn(3, 8, 16)

ARGS = {
    'data': data,
    'dimOrder': DimOrder.ChannelHeightWidth,
    'maxWindowSizeW': 4,
    'maxWindowSizeH': 4,
    'overlapPercent': 0.25,
    'transforms': [simple_transform]
}
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 list, ndim 1, dtype family float — a list 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
generateForSize Ran
ran on a synthesized input
ibm/max-human-pose-estimator/core/tf_pose/slidingwindow/SlidingWindow.py
code served (permissive licence) · get_code("93c5229a4050e760")
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 list, ndim 1, 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 math

class DimOrder(object):
    """
    Represents the order of the dimensions in a dataset's shape.
    """
    ChannelHeightWidth = ['c', 'h', 'w']
    HeightWidthChannel = ['h', 'w', 'c']

class SlidingWindow(object):
    """
    Represents a single window into a larger dataset.
    """

    def __init__(self, x, y, w, h, dimOrder, transform=None):
        """
        Creates a new window with the specified dimensions and transform
        """
        self.x = x
        self.y = y
        self.w = w
        self.h = h
        self.dimOrder = dimOrder
        self.transform = transform

    def apply(self, matrix):
        """
        Slices the supplied matrix and applies any transform bound to this window
        """
        view = matrix[self.indices()]
        return self.transform(view) if self.transform is not None else view

    def getRect(self):
        """
        Returns the window bounds as a tuple of (x,y,w,h)
        """
        return (self.x, self.y, self.w, self.h)

    def setRect(self, rect):
        """
        Sets the window bounds from a tuple of (x,y,w,h)
        """
        self.x, self.y, self.w, self.h = rect

    def indices(self, includeChannel=True):
        """
        Retrieves the indices for this window as a tuple of slices
        """
        if self.dimOrder == DimOrder.HeightWidthChannel:
            return (slice(self.y, self.y + self.h), slice(self.x, self.x + self.w))
        elif self.dimOrder == DimOrder.ChannelHeightWidth:
            if includeChannel is True:
                return (slice(None, None), slice(self.y, self.y + self.h), slice(self.x, self.x + self.w))
            else:
                return (slice(self.y, self.y + self.h), slice(self.x, self.x + self.w))
        else:
            raise ValueError('Unsupported order of dimensions: ' + str(self.dimOrder))

    def __str__(self):
        return '(' + str(self.x) + ',' + str(self.y) + ',' + str(self.w) + ',' + str(self.h) + ')'

    def __repr__(self):
        return self.__str__()

def generateForSize(width, height, dimOrder, maxWindowSizeW, maxWindowSizeH, overlapPercent, transforms=[]):
    """
    Generates a set of sliding windows for a dataset with the specified dimensions and order.
    """
    windowSizeX = min(maxWindowSizeW, width)
    windowSizeY = min(maxWindowSizeH, height)
    windowOverlapX = int(math.floor(windowSizeX * overlapPercent))
    windowOverlapY = int(math.floor(windowSizeY * overlapPercent))
    stepSizeX = windowSizeX - windowOverlapX
    stepSizeY = windowSizeY - windowOverlapY
    lastX = width - windowSizeX
    lastY = height - windowSizeY
    xOffsets = list(range(0, lastX + 1, stepSizeX))
    yOffsets = list(range(0, lastY + 1, stepSizeY))
    if len(xOffsets) == 0 or xOffsets[-1] != lastX:
        xOffsets.append(lastX)
    if len(yOffsets) == 0 or yOffsets[-1] != lastY:
        yOffsets.append(lastY)
    windows = []
    for xOffset in xOffsets:
        for yOffset in yOffsets:
            for transform in [None] + transforms:
                windows.append(SlidingWindow(x=xOffset, y=yOffset, w=windowSizeX, h=windowSizeY, dimOrder=dimOrder, transform=transform))
    return windows

def simple_transform(x):
    return x * 2

ARGS = {
    'width': 64,
    'height': 48,
    'dimOrder': DimOrder.ChannelHeightWidth,
    'maxWindowSizeW': 32,
    'maxWindowSizeH': 24,
    'overlapPercent': 0.25,
    'transforms': [simple_transform]
}
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 list, ndim 1, dtype family float — a list 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
get_sample_images Ran
ran on a synthesized input
ibm/max-human-pose-estimator/core/tf_pose/common.py
code served (permissive licence) · get_code("1cb22812d63a8d83")
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 list, ndim 4, 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 cv2
import numpy as np
import tempfile
import os

# Create a temporary directory and a simple test image
temp_dir = tempfile.mkdtemp()
image_path = os.path.join(temp_dir, 'Pilots.jpg')

# Create a minimal valid JPEG image (100x100 RGB)
dummy_image = np.random.randint(0, 256, (100, 100, 3), dtype=np.uint8)
cv2.imwrite(image_path, dummy_image)

# Mock the './images/Pilots.jpg' path by monkey-patching cv2.imread
original_imread = cv2.imread

def mock_imread(path, flags=cv2.IMREAD_COLOR):
    if 'Pilots.jpg' in path:
        return original_imread(image_path, flags)
    return original_imread(path, flags)

cv2.imread = mock_imread

ARGS = {
    'w': 640,
    'h': 480
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 4, dtype family int — a list 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
layer Ran
ran; our contract draft was wrong, not the code
ibm/max-human-pose-estimator/core/tf_pose/tensblur/smoother.py
code served (permissive licence) · get_code("1fb38eb9d6522fb0")
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 torch
import torch.nn as nn

torch.manual_seed(0)

class MockLayer(nn.Module):
    def __init__(self):
        super().__init__()
        self.terminals = []
        self.layers = {}
        self._name_counter = {}
    
    def get_unique_name(self, name):
        if name not in self._name_counter:
            self._name_counter[name] = 0
        else:
            self._name_counter[name] += 1
        return f"{name}_{self._name_counter[name]}"
    
    def feed(self, output):
        self.terminals = [output]

def mock_op(self, layer_input, *args, **kwargs):
    """A simple mock operation that returns the input or processes it"""
    if isinstance(layer_input, list):
        return torch.cat(layer_input, dim=1)
    else:
        return layer_input * 2.0

mock_instance = MockLayer()
mock_instance.terminals = [torch.randn(2, 4)]

ARGS = {
    'op': mock_op
}
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 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
model_wh Ran
ran on a synthesized input
ibm/max-human-pose-estimator/core/tf_pose/networks.py
code served (permissive licence) · get_code("f8b79dbb86b46205")
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 torch
import numpy as np

torch.manual_seed(0)

# The function model_wh takes a resolution_str parameter
# It parses a string in format "WIDTHxHEIGHT" and validates that both are multiples of 16
# Then returns a tuple of (width, height) as integers

# Valid resolution string with dimensions that are multiples of 16
resolution_str = "320x240"

ARGS = {
    "resolution_str": resolution_str
}
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 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
normalize Ran
ran, violated the contract we drafted
Xingyu-Romantic/RMPose_PAF/testing/test_pose.py
pointer only (licence: NONE) · get_code("9d8d1b763badb5e7")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, Xingyu-Romantic/RMPose_PAF 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.

padRect Ran
ran on a synthesized input
ibm/max-human-pose-estimator/core/tf_pose/slidingwindow/RectangleUtils.py
code served (permissive licence) · get_code("d8dbe9ff57d30fbc")
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 torch
import numpy as np

# Analyze the function:
# - rect: tuple of (x, y, w, h) - rectangle coordinates and dimensions
# - padTop, padBottom, padLeft, padRight: integers for padding amounts
# - bounds: tuple with at least 2 elements (bounds[0] and bounds[1])
# - clipExcess: boolean flag

# The function modifies a rectangle by padding it and clipping/adjusting to bounds
# bounds appears to be (height, width) based on the logic:
#   overflowY = max(0, y + h - bounds[0])  # compared against bounds[0]
#   overflowX = max(0, x + w - bounds[1])  # compared against bounds[1]

ARGS = {
    "rect": (10, 20, 50, 60),  # (x, y, w, h) - a rectangle at position (10,20) with width 50, height 60
    "padTop": 5,
    "padBottom": 5,
    "padLeft": 3,
    "padRight": 3,
    "bounds": (100, 100),  # (max_height, max_width) - the bounding area
    "clipExcess": True
}
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 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
padRightDownCorner Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Xingyu-Romantic/RMPose_PAF/testing/test_pose.py
pointer only (licence: NONE) · get_code("1273c2cd8619f6cb")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, Xingyu-Romantic/RMPose_PAF 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.

plot_result Ran
ran on a synthesized input
MSeeker1340/Vision2018-Pose/src/plot.py
code served (permissive licence) · get_code("822b0c684ba9eba6")
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 3, dtype family float, shape (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

# Create a sample 3D array with shape (height, width, channels)
# The function will extract channel f and resize it to (224, 224)
y = np.random.rand(128, 128, 5).astype(np.float32)

# f is the channel index to extract
f = 2

ARGS = {
    "y": y,
    "f": f,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, skimage — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family float, shape (224, 224, 3) — shape (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
read_imgfile Ran
ran on a synthesized input
ibm/max-human-pose-estimator/core/tf_pose/common.py
code served (permissive licence) · get_code("1440a7103327c0de")
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 3, dtype family int, shape (6, 8, 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 tempfile
import cv2
import numpy as np

# Create a temporary image file for testing
with tempfile.NamedTemporaryFile(suffix='.jpg', delete=False) as tmp:
    # Create a small test image (3x4 RGB)
    test_image = np.random.randint(0, 256, (3, 4, 3), dtype=np.uint8)
    cv2.imwrite(tmp.name, test_image)
    temp_path = tmp.name

ARGS = {
    'path': temp_path,
    'width': 8,
    'height': 6
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family int, shape (6, 8, 3) — shape (6, 8, 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
round_int Ran
ran, honoured the contract we drafted
DavHoffmann/LearningToTrainWithSyntheticHumans/tf_pose/eval_MPI.py
pointer only (licence: NOASSERTION) · get_code("244255938544c7a6")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, DavHoffmann/LearningToTrainWithSyntheticHumans (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.

zerosFactory Ran
ran on a synthesized input
ibm/max-human-pose-estimator/core/tf_pose/slidingwindow/ArrayUtils.py
code served (permissive licence) · get_code("e25580edeb9875ea")
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 3, dtype family float, shape (4, 5, 6) 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
import mmap
import tempfile
import numpy as np
import psutil

def _requiredSize(shape, dtype):
    """
    Determines the number of bytes required to store a NumPy array with
    the specified shape and datatype.
    """
    return math.floor(np.prod(np.asarray(shape, dtype=np.uint64)) * np.dtype(dtype).itemsize)

class TempfileBackedArray(np.ndarray):
    """
    A NumPy ndarray that uses a memory-mapped temp file as its backing
    """

    def __new__(subtype, shape, dtype=float, buffer=None, offset=0, strides=None, order=None, info=None):
        numBytes = _requiredSize(shape, dtype)
        tempFile = tempfile.TemporaryFile()
        tempFile.truncate(numBytes)
        buf = mmap.mmap(tempFile.fileno(), numBytes, access=mmap.ACCESS_WRITE)
        obj = super(TempfileBackedArray, subtype).__new__(subtype, shape, dtype, buf, 0, None, order)
        obj._file = tempFile
        return obj

    def __array_finalize__(self, obj):
        if obj is None:
            return
        self._file = getattr(obj, '_file', None)

def arrayFactory(shape, dtype=float):
    """
    Creates a new ndarray of the specified shape and datatype, storing
    it in memory if there is sufficient available space or else using
    a memory-mapped temporary file to provide the underlying buffer.
    """
    requiredBytes = _requiredSize(shape, dtype)
    vmem = psutil.virtual_memory()
    if vmem.available > requiredBytes:
        return np.ndarray(shape=shape, dtype=dtype)
    else:
        return TempfileBackedArray(shape=shape, dtype=dtype)

def zerosFactory(shape, dtype=float):
    """
    Creates a new NumPy array using `arrayFactory()` and fills it with zeros.
    """
    arr = arrayFactory(shape=shape, dtype=dtype)
    arr.fill(0)
    return arr

ARGS = {
    "shape": (4, 5, 6),
    "dtype": np.float32
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, psutil — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family float, shape (4, 5, 6) — shape (4, 5, 6) 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
generateDistanceMatrix Not yet run
unverified (harvested, no recorded run)
ibm/max-human-pose-estimator/core/tf_pose/slidingwindow/WindowDistance.py
code served (permissive licence) · get_code("1fc7d25ab7ceeb40")
How it ran

no invocation recorded.

get_part_cid Not yet run
unverified (harvested, no recorded run)
ibm/max-human-pose-estimator/core/tf_pose/pafprocess/pafprocess.py
code served (permissive licence) · get_code("0ad190b8ea7cf881")
How it ran

no invocation recorded.

get_score Not yet run
unverified (harvested, no recorded run)
ibm/max-human-pose-estimator/core/tf_pose/pafprocess/pafprocess.py
code served (permissive licence) · get_code("418d24f644276625")
How it ran

no invocation recorded.

mergeWindows Not yet run
unverified (harvested, no recorded run)
ibm/max-human-pose-estimator/core/tf_pose/slidingwindow/Merging.py
code served (permissive licence) · get_code("33f87ef10f320887")
How it ran

no invocation recorded.

process_paf Not yet run
unverified (harvested, no recorded run)
ibm/max-human-pose-estimator/core/tf_pose/pafprocess/pafprocess.py
code served (permissive licence) · get_code("dd2c2980939419d2")
How it ran

no invocation recorded.

Also present in other papers' code

Of 23 samples lifted from this paper's linked repositories, 23 carry a function identity and 0 do not. Of the 23 distinct functions with an identity, 2 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 2 later) and 21 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
layer 5 other papers
1 dated earlier, 4 later
1606.00373 (2016-06)
Deeper Depth Prediction with Fully Convolutional Residual Networks
1808.00157 (2018-08)
Instance-level Human Parsing via Part Grouping Network
1903.08527 (2019-03)
Accurate 3D Face Reconstruction with Weakly-Supervised Learning: From Single Image to Image Set
and 2 more not named here
padRightDownCorner 2 other papers
2 later
2307.15880 (2023-07)
Effective Whole-body Pose Estimation with Two-stages Distillation
2410.24037 (2024-10)
TPC: Test-time Procrustes Calibration for Diffusion-based Human Image Animation

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 2026-11-21.

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 an approach to efficiently detect the 2D pose of multiple people in an image. The approach uses a nonparametric representation, which we refer to as Part Affinity Fields (PAFs), to learn to associate body parts with individuals in the image. The architecture encodes global context, allowing a greedy bottom-up parsing step that maintains high accuracy while achieving realtime performance, irrespective of the number of people in the image. The architecture is designed to jointly learn part locations and their association via two branches of the same sequential prediction process. Our method placed first in the inaugural COCO 2016 keypoints challenge, and significantly exceeds the previous state-of-the-art result on the MPII Multi-Person benchmark, both in performance and efficiency.

For agents

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

get_harvested_code_for_paper("1611.08050")
get_code_for_paper("1611.08050")
have("1611.08050")

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

curl https://syntology.ai/api/ran/1611.08050.json

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

[![Syntology run record](https://syntology.ai/api/ran/1611.08050.svg)](https://syntology.ai/paper/1611.08050)

Connect an agent — have() is free.