SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2003.03771 · 2020

Pixel-in-Pixel Net: Towards Efficient Facial Landmark Detection in the Wild

arXiv · PDF · Open in the Atlas

Code that ran

15 of 17 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; 0 honoured the contract we drafted, 0 violated it and 13 have no contract verdict; 2 failures of our instrument; 2 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/2003.03771.json · badge: /api/ran/2003.03771.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
jhb86253817/PIPNet canonical (classifier label)
classifier basis: census
13 ran · 2 no recorded run
deftruth/torchlm community (classifier label)
classifier basis not recorded on the link
2 ran · 0 no recorded run
FunctionStatusWhere it lives
compute_loss_pip Ran
ran on a synthesized input
jhb86253817/PIPNet/lib/functions_gssl.py
code served (permissive licence) · get_code("0ecee63670b4125b")
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 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

torch.manual_seed(0)

# Define small criterion objects that work like loss functions
class DummyClassificationLoss(nn.Module):
    def forward(self, output, target):
        return torch.mean((output - target) ** 2)

class DummyRegressionLoss(nn.Module):
    def forward(self, output, target):
        return torch.mean((output - target) ** 2)

# Set up dimensions
batch_size = 2
num_channels = 4
height = 3
width = 5
num_nb = 2  # number of neighbors
spatial_size = height * width

# Create output tensors for maps (batch, channel, height, width)
outputs_map1 = torch.randn(batch_size, num_channels, height, width)
outputs_map2 = torch.randn(batch_size, num_channels, height, width)
outputs_map3 = torch.randn(batch_size, num_channels, height, width)

# Create label tensors for maps
labels_map1 = torch.randn(batch_size, num_channels, height, width)
labels_map2 = torch.randn(batch_size, num_channels, height, width)
labels_map3 = torch.randn(batch_size, num_channels, height, width)

# Create mask tensors for maps
masks_map1 = torch.randint(0, 2, (batch_size, num_channels, height, width)).float()
masks_map2 = torch.randint(0, 2, (batch_size, num_channels, height, width)).float()
masks_map3 = torch.randint(0, 2, (batch_size, num_channels, height, width)).float()

# Create local output tensors (batch, channel, spatial_size)
outputs_local_x = torch.randn(batch_size, num_channels, spatial_size)
outputs_local_y = torch.randn(batch_size, num_channels, spatial_size)

# Create local label tensors
labels_local_x = torch.randn(batch_size, num_channels, spatial_size)
labels_local_y = torch.randn(batch_size, num_channels, spatial_size)

# Create local mask tensors
masks_local_x = torch.randint(0, 2, (batch_size, num_channels, spatial_size)).float()
masks_local_y = torch.randint(0, 2, (batch_size, num_channels, spatial_size)).float()

# Create neighbor output tensors (batch, num_nb, channel, spatial_size)
outputs_nb_x = torch.randn(batch_size, num_nb, num_channels, spatial_size)
outputs_nb_y = torch.randn(batch_size, num_nb, num_channels, spatial_size)

# Create neighbor label tensors
labels_nb_x = torch.randn(batch_size, num_nb, num_channels, spatial_size)
labels_nb_y = torch.randn(batch_size, num_nb, num_channels, spatial_size)

# Create neighbor mask tensors
masks_nb_x = torch.randint(0, 2, (batch_size, num_nb, num_channels, spatial_size)).float()
masks_nb_y = torch.randint(0, 2, (batch_size, num_nb, num_channels, spatial_size)).float()

# Create criterion functions
criterion_cls = DummyClassificationLoss()
criterion_reg = DummyRegressionLoss()

ARGS = {
    'outputs_map1': outputs_map1,
    'outputs_map2': outputs_map2,
    'outputs_map3': outputs_map3,
    'outputs_local_x': outputs_local_x,
    'outputs_local_y': outputs_local_y,
    'outputs_nb_x': outputs_nb_x,
    'outputs_nb_y': outputs_nb_y,
    'labels_map1': labels_map1,
    'labels_map2': labels_map2,
    'labels_map3': labels_map3,
    'labels_local_x': labels_local_x,
    'labels_local_y': labels_local_y,
    'labels_nb_x': labels_nb_x,
    'labels_nb_y': labels_nb_y,
    'masks_map1': masks_map1,
    'masks_map2': masks_map2,
    'masks_map3': masks_map3,
    'masks_local_x': masks_local_x,
    'masks_local_y': masks_local_y,
    'masks_nb_x': masks_nb_x,
    'masks_nb_y': masks_nb_y,
    'criterion_cls': criterion_cls,
    'criterion_reg': criterion_reg,
    'num_nb': num_nb,
}
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 float — 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
conv_1x1_bn Ran
ran; our contract draft was wrong, not the code
jhb86253817/PIPNet/lib/mobilenetv3.py
code served (permissive licence) · get_code("4e90491a30c2dc28")
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)

# Define the custom activation modules needed by conv_1x1_bn
class h_sigmoid(nn.Module):
    def __init__(self, inplace=True):
        super(h_sigmoid, self).__init__()
        self.relu = nn.ReLU6(inplace=inplace)

    def forward(self, x):
        return self.relu(x + 3) / 6

class h_swish(nn.Module):
    def __init__(self, inplace=True):
        super(h_swish, self).__init__()
        self.sigmoid = h_sigmoid(inplace=inplace)

    def forward(self, x):
        return x * self.sigmoid(x)

def conv_1x1_bn(inp, oup):
    return nn.Sequential(
        nn.Conv2d(inp, oup, 1, 1, 0, bias=False),
        nn.BatchNorm2d(oup),
        h_swish()
    )

# Create arguments for conv_1x1_bn
# inp: number of input channels (distinct from oup)
# oup: number of output channels
inp_channels = 4
oup_channels = 3

ARGS = {
    'inp': inp_channels,
    'oup': oup_channels
}
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
conv_3x3_bn Ran
ran; our contract draft was wrong, not the code
jhb86253817/PIPNet/lib/mobilenetv3.py
code served (permissive licence) · get_code("5221887077e3fb54")
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': 3,
    'oup': 16,
    '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
fr_and_auc Ran
ran on a synthesized input
deftruth/torchlm/torchlm/metrics/metrics.py
code served (permissive licence) · get_code("7a33e0cbdc0a9b10")
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 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 numpy as np

# fr_and_auc takes nmes (list of floats) and optional thres and step parameters
# The function counts how many nmes values are <= each threshold in xs
# and computes a failure rate and AUC

ARGS = {
    "nmes": [0.05, 0.08, 0.12, 0.15, 0.02, 0.11],
    "thres": 0.1,
    "step": 0.0001,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — 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 float — 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
get_label Ran
ran on a synthesized input
jhb86253817/PIPNet/lib/functions.py
code served (permissive licence) · get_code("c6f9712584e0d9f8")
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. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import tempfile
import numpy as np

# Create a temporary directory structure with test label files
temp_dir = tempfile.mkdtemp()
data_dir = os.path.join(temp_dir, 'data', 'test_dataset')
os.makedirs(data_dir, exist_ok=True)

# Create a label file with image names and targets
label_file_path = os.path.join(data_dir, 'labels.txt')
with open(label_file_path, 'w') as f:
    f.write('image1.jpg 0.1 0.2 0.3\n')
    f.write('image2.jpg 0.4 0.5 0.6\n')
    f.write('image3.jpg 0.7 0.8 0.9\n')

# Change to the temp directory so relative paths work
original_dir = os.getcwd()
os.chdir(temp_dir)

ARGS = {
    'data_name': 'test_dataset',
    'label_file': 'labels.txt',
    'task_type': None
}
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 — 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
get_meanface Ran
ran on a synthesized input
jhb86253817/PIPNet/lib/functions.py
code served (permissive licence) · get_code("d4eeabbb4d3bfbcb")
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. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import tempfile
import os

# Create a temporary meanface file
temp_dir = tempfile.mkdtemp()
meanface_file = os.path.join(temp_dir, "meanface.txt")

# Create sample meanface data with 5 points in 2D (x, y coordinates)
# 5 points means 10 floats total
meanface_data = "0.0 0.0 1.0 0.0 2.0 0.0 3.0 1.0 1.5 2.0"
with open(meanface_file, 'w') as f:
    f.write(meanface_data)

ARGS = {
    "meanface_file": meanface_file,
    "num_nb": 2
}
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 tuple — a tuple 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
nme Ran
ran on a synthesized input
deftruth/torchlm/torchlm/metrics/metrics.py
code served (permissive licence) · get_code("791cd5847cdd6ff5")
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 float32, ndim 0, dtype family float, 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 numpy as np

# Create small sample data for landmarks
# Using distinct dimensions: 5 landmarks with 2 coordinates each
n_landmarks = 5

# Predicted landmarks: (5, 2)
lms_pred = np.array([
    [10.0, 20.0],
    [15.5, 25.3],
    [12.1, 22.8],
    [18.2, 28.1],
    [11.9, 21.5]
], dtype=np.float32)

# Ground truth landmarks: (5, 2)
lms_gt = np.array([
    [10.2, 20.1],
    [15.3, 25.0],
    [12.0, 22.9],
    [18.0, 28.2],
    [12.1, 21.4]
], dtype=np.float32)

# Normalize value (distance between eyeballs, typically around 100 pixels)
norm = 100.0

ARGS = {
    "lms_pred": lms_pred,
    "lms_gt": lms_gt,
    "norm": norm
}
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 float32, ndim 0, dtype family float, 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
process_300w Ran
ran on a synthesized input
jhb86253817/PIPNet/lib/preprocess.py
code served (permissive licence) · get_code("22d83c9db600623b")
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. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import cv2
import numpy as np
import tempfile

# Create a temporary directory structure with test data
temp_dir = tempfile.mkdtemp()
test_folder = os.path.join(temp_dir, "test_folder")
os.makedirs(test_folder, exist_ok=True)

# Create a test image (100x100 BGR image)
test_image = np.random.randint(0, 255, (100, 100, 3), dtype=np.uint8)
image_path = os.path.join(test_folder, "test_image.jpg")
cv2.imwrite(image_path, test_image)

# Create a test label file with landmark annotations
# Format: header lines, then landmark coordinates, then footer
label_content = """version: 1
npoints: 5
{
10 20
30 40
50 60
70 80
90 90
}
"""
label_path = os.path.join(test_folder, "test_label.txt")
with open(label_path, 'w') as f:
    f.write(label_content)

ARGS = {
    "root_folder": temp_dir,
    "folder_name": "test_folder",
    "image_name": "test_image.jpg",
    "label_name": "test_label.txt",
    "target_size": 64
}
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 tuple — a tuple 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
process_300w Ran
ran on a synthesized input
jhb86253817/PIPNet/lib/preprocess_gssl.py
code served (permissive licence) · get_code("5a75fb8671b68880")
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. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import tempfile
import cv2
import numpy as np

# Create temporary directory structure with test files
temp_dir = tempfile.mkdtemp()
folder_name = "test_folder"
folder_path = os.path.join(temp_dir, folder_name)
os.makedirs(folder_path, exist_ok=True)

# Create a test image (100x120 BGR)
image_array = np.random.randint(0, 255, (120, 100, 3), dtype=np.uint8)
image_name = "test_image.jpg"
image_path = os.path.join(folder_path, image_name)
cv2.imwrite(image_path, image_array)

# Create a test label file with landmark annotations
# Format: first 3 lines are headers, then landmark coordinates, then some footer
label_name = "test_image.pts"
label_path = os.path.join(folder_path, label_name)
with open(label_path, 'w') as f:
    f.write("version: 1\n")
    f.write("n_points: 68\n")
    f.write("{\n")
    # Write 68 landmark points (x y coordinates)
    for i in range(68):
        x = 20 + (i % 10) * 5
        y = 30 + (i // 10) * 5
        f.write(f"{x} {y}\n")
    f.write("}\n")

ARGS = {
    "root_folder": temp_dir,
    "folder_name": folder_name,
    "image_name": image_name,
    "label_name": label_name,
    "target_size": 64
}
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 tuple — a tuple 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
process_celeba Ran
ran on a synthesized input
jhb86253817/PIPNet/lib/preprocess_gssl.py
code served (permissive licence) · get_code("657f44be85c2cb01")
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 (128, 128, 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 os
import cv2
import numpy as np
import tempfile

# Create a temporary directory structure that mimics CelebA
temp_dir = tempfile.mkdtemp()
celeba_dir = os.path.join(temp_dir, 'CELEBA', 'img_celeba')
os.makedirs(celeba_dir, exist_ok=True)

# Create a small dummy image file
image_name = 'test_image.jpg'
dummy_image = np.random.randint(0, 256, (256, 256, 3), dtype=np.uint8)
image_path = os.path.join(celeba_dir, image_name)
cv2.imwrite(image_path, dummy_image)

# Define the bounding box coordinates (xmin, ymin, xmax, ymax)
bbox = (50, 60, 150, 180)

# Define target size for resizing
target_size = 128

ARGS = {
    'root_folder': temp_dir,
    'image_name': image_name,
    'bbox': bbox,
    'target_size': target_size
}
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 (128, 128, 3) — shape (128, 128, 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
process_cofw Ran
ran on a synthesized input
jhb86253817/PIPNet/lib/preprocess.py
code served (permissive licence) · get_code("ac5ae0ae9d4b74aa")
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. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import cv2
import numpy as np

# Create a small valid image
image = np.random.randint(0, 256, size=(100, 120, 3), dtype=np.uint8)

# Create bounding box as (xmin, ymin, width, height)
bbox = (10, 15, 50, 60)

# Create annotation: 29 x-coordinates followed by 29 y-coordinates
anno_x = np.random.uniform(10, 60, size=29)
anno_y = np.random.uniform(15, 75, size=29)
anno = np.concatenate([anno_x, anno_y])

# Target size for resizing
target_size = 64

ARGS = {
    "image": image,
    "bbox": bbox,
    "anno": anno,
    "target_size": target_size,
}
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 tuple — a tuple 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
process_wflw Ran
ran on a synthesized input
jhb86253817/PIPNet/lib/preprocess_gssl.py
code served (permissive licence) · get_code("149fe226ec8dbea0")
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. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import cv2
import numpy as np
from unittest.mock import Mock, patch

# Create a minimal mock image as a numpy array
# The function reads an image, so we need to mock cv2.imread
mock_image = np.random.randint(0, 255, (480, 640, 3), dtype=np.uint8)

# Create annotation data: 196 landmark coordinates + 4 bbox values + 6 attributes + 1 image name
# 196 landmarks (98 points * 2 for x,y)
lms_coords = [float(i % 640) for i in range(196)]  # 98 x,y pairs within image bounds

# 4 bbox values: xmin, ymin, xmax, ymax
bbox_values = [100.0, 120.0, 500.0, 450.0]

# 6 attribute values (0 or 1)
attrs = ['1', '0', '1', '0', '1', '0']

# Image name
image_name = 'test_image.jpg'

# Combine all into annotation list
anno = lms_coords + bbox_values + attrs + [image_name]

# Mock cv2.imread to return our mock image
def mock_imread(path):
    return mock_image

# Target size for resizing
target_size = 224

ARGS = {
    'anno': anno,
    'target_size': target_size
}

# Patch cv2.imread for the function execution
import sys
cv2.imread = mock_imread
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 tuple — a tuple 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
random_blur Ran
ran on a synthesized input
jhb86253817/PIPNet/lib/data_utils.py
code served (permissive licence) · get_code("4e4f39068434b6e0")
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 Image. 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.

from PIL import Image, ImageFilter
import random

# Create a small PIL Image object
# The function calls .filter() method on the image
image = Image.new('RGB', (32, 32), color='red')

ARGS = {
    'image': image
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Image — a Image output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Image output the recorded ndim 3 and dtype family int 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
random_occlusion Ran
ran on a synthesized input
jhb86253817/PIPNet/lib/data_utils.py
code served (permissive licence) · get_code("47e71b4e99e8b661")
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 Image. 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.

from PIL import Image
import numpy as np
import random

random.seed(42)

# Create a small test image (RGB format)
image_array = np.random.randint(0, 256, size=(100, 80, 3), dtype=np.uint8)
image = Image.fromarray(image_array, 'RGB')

ARGS = {
    'image': image
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Image — a Image output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Image output the recorded ndim 3 and dtype family int 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
random_translate Ran
ran on a synthesized input
jhb86253817/PIPNet/lib/data_utils.py
code served (permissive licence) · get_code("555d941276e9070f")
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. 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 numpy as np
from PIL import Image
import random

# Set seed for reproducibility
random.seed(0)
torch.manual_seed(0)

# Create a small PIL Image (RGB mode, 100x100 pixels)
image = Image.new('RGB', (100, 100), color=(255, 0, 0))

# Create a target array representing normalized coordinates
# Shape: (4,) - representing 2 points with (x, y) normalized coordinates
target = np.array([0.3, 0.4, 0.7, 0.8], dtype=np.float32)

ARGS = {
    'image': image,
    'target': target
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple 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
compute_loss_pip Not yet run
unverified (harvested, no recorded run)
jhb86253817/PIPNet/lib/functions.py
code served (permissive licence) · get_code("180321558d57e253")
How it ran

no invocation recorded.

process_wflw Not yet run
unverified (harvested, no recorded run)
jhb86253817/PIPNet/lib/preprocess.py
code served (permissive licence) · get_code("260fe0df73217f72")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 distinct functions with an identity, 2 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 2 later) and 15 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_1x1_bn 4 other papers
2 dated earlier, 2 later
1512.02325 (2015-12)
SSD: Single Shot MultiBox Detector
1905.02244 (2019-05)
Searching for MobileNetV3
Wang_3D_Face_Reconstruction_with_the_Geometric_Guidance_of_Facial_Part_CVPR_2024_paper (2024 (year only))
3D Face Reconstruction with the Geometric Guidance of Facial Part Segmentation
and 1 more not named here
conv_3x3_bn 4 other papers
2 dated earlier, 2 later
1512.02325 (2015-12)
SSD: Single Shot MultiBox Detector
1905.02244 (2019-05)
Searching for MobileNetV3
Wang_3D_Face_Reconstruction_with_the_Geometric_Guidance_of_Facial_Part_CVPR_2024_paper (2024 (year only))
3D Face Reconstruction with the Geometric Guidance of Facial Part Segmentation
and 1 more not named here

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

Repositories linked to this paper

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

Abstract

Recently, heatmap regression models have become popular due to their superior performance in locating facial landmarks. However, three major problems still exist among these models: (1) they are computationally expensive; (2) they usually lack explicit constraints on global shapes; (3) domain gaps are commonly present. To address these problems, we propose Pixel-in-Pixel Net (PIPNet) for facial landmark detection. The proposed model is equipped with a novel detection head based on heatmap regression, which conducts score and offset predictions simultaneously on low-resolution feature maps. By doing so, repeated upsampling layers are no longer necessary, enabling the inference time to be largely reduced without sacrificing model accuracy. Besides, a simple but effective neighbor regression module is proposed to enforce local constraints by fusing predictions from neighboring landmarks, which enhances the robustness of the new detection head. To further improve the cross-domain generalization capability of PIPNet, we propose self-training with curriculum. This training strategy is able to mine more reliable pseudo-labels from unlabeled data across domains by starting with an easier task, then gradually increasing the difficulty to provide more precise labels. Extensive experiments demonstrate the superiority of PIPNet, which obtains state-of-the-art results on three out of six popular benchmarks under the supervised setting. The results on two cross-domain test sets are also consistently improved compared to the baselines. Notably, our lightweight version of PIPNet runs at 35.7 FPS and 200 FPS on CPU and GPU, respectively, while still maintaining a competitive accuracy to state-of-the-art methods. The code of PIPNet is available at https://github.com/jhb86253817/PIPNet.

For agents

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

get_harvested_code_for_paper("2003.03771")
get_code_for_paper("2003.03771")
have("2003.03771")

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

curl https://syntology.ai/api/ran/2003.03771.json

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

[![Syntology run record](https://syntology.ai/api/ran/2003.03771.svg)](https://syntology.ai/paper/2003.03771)

Connect an agent — have() is free.