SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2303.14736 · CVPR · 2023

Disentangling Writer and Character Styles for Handwriting Generation

Yifan Zhang, Zhuoman Liu, Zhuliang Yu, Shuangping Huang, Qing Du, Gang Dai, Qingfeng Wang

arXiv · PDF · Open in the Atlas

Code that ran

14 of 16 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; 1 failure 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/2303.14736.json · badge: /api/ran/2303.14736.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
dailenson/SDT canonical (classifier label)
classifier basis not recorded on the link
8 ran · 2 no recorded run
koninik/htg_evaluation community (classifier label)
classifier basis not recorded on the link
6 ran · 0 no recorded run
FunctionStatusWhere it lives
affine_transformation Ran
ran on a synthesized input
koninik/htg_evaluation/utils/auxilary_functions.py
code served (permissive licence) · get_code("22a46289db28df6d")
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 int, shape (4, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import cv2

# Create a small test image
# The function expects img to have shape (h, w) or (h, w, c)
# Using a small 4x3 grayscale image for simplicity
img = np.random.randint(0, 256, size=(4, 3), dtype=np.uint8)

ARGS = {
    "img": img,
    "m": 1.0,
    "s": 0.2,
    "border_value": 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 ndarray, ndim 2, dtype family int, shape (4, 3) — shape (4, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
centered Ran
ran on a synthesized input
koninik/htg_evaluation/utils/auxilary_functions.py
code served (permissive licence) · get_code("ae6d12745b44744a")
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 int, shape (6, 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 numpy as np

# Create a small word image (4x3 array)
word_img = np.array([
    [100, 150, 200],
    [120, 160, 210],
    [110, 155, 205],
    [130, 170, 220]
], dtype=np.uint8)

# Target size (height, width)
tsize = (6, 5)

# Centering tuple
centering = (0.5, 0.5)

# Border value
border_value = None

ARGS = {
    "word_img": word_img,
    "tsize": tsize,
    "centering": centering,
    "border_value": border_value
}
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 ndarray, ndim 2, dtype family int, shape (6, 5) — shape (6, 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
eval_class_epoch Ran
ran on a synthesized input
koninik/htg_evaluation/htg_style.py
code served (permissive licence) · get_code("271b9e1e885936bf")
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)

# Create a minimal model that takes an image tensor and outputs class logits
class MinimalModel(nn.Module):
    def __init__(self, num_classes=5):
        super().__init__()
        self.fc = nn.Linear(4, num_classes)
    
    def forward(self, x):
        # Flatten input and pass through linear layer
        x = x.view(x.size(0), -1)
        return self.fc(x)

# Create minimal validation data - list of tuples (image, label1, label2)
# The function uses data[0] for image and data[2] for label
validation_data = [
    (torch.randn(2, 4), torch.randint(0, 5, (2,)), torch.randint(0, 5, (2,))),
    (torch.randn(2, 4), torch.randint(0, 5, (2,)), torch.randint(0, 5, (2,))),
]

# Create args object with required attributes
class Args:
    def __init__(self):
        self.dataset = 'not_iam'  # Will use data[1], but code overrides to data[2]
        self.device = 'cpu'

model = MinimalModel(num_classes=5)
args = Args()

ARGS = {
    'model': model,
    'validation_data': validation_data,
    'args': args
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, tqdm — 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_mixture_coef Ran
ran on a synthesized input
dailenson/SDT/models/gmm.py
code served (permissive licence) · get_code("3938029b16a8e652")
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. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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

torch.manual_seed(0)

# The function expects output to be a tensor
# It splits it as: z[:, 0:3] for pen_logits, then z[:, 3:] for the rest
# z[:, 3:] is split into 6 tensors of size 20 each
# So total width needed: 3 + 6*20 = 123
# Batch size: 2

output = torch.randn(2, 123)

ARGS = {
    "output": output
}
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 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_pen_loss Ran
ran on a synthesized input
dailenson/SDT/models/loss.py
code served (permissive licence) · get_code("c3c5ed7bdf6dff50")
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

torch.manual_seed(0)

# Batch size: 2, number of mixture components: 4, pen_data has 3 columns
batch_size = 2
num_components = 4

# z_pi: mixture weights for each component, shape (batch_size, num_components)
z_pi = torch.softmax(torch.randn(batch_size, num_components), dim=1)

# z_mu1, z_mu2: means for x1 and x2, shape (batch_size, num_components)
z_mu1 = torch.randn(batch_size, num_components)
z_mu2 = torch.randn(batch_size, num_components)

# z_sigma1, z_sigma2: standard deviations, shape (batch_size, num_components)
z_sigma1 = torch.abs(torch.randn(batch_size, num_components)) + 0.1
z_sigma2 = torch.abs(torch.randn(batch_size, num_components)) + 0.1

# z_corr: correlation coefficient, shape (batch_size, num_components)
z_corr = torch.tanh(torch.randn(batch_size, num_components))

# z_pen_logits: logits for pen classification, shape (batch_size, num_classes)
num_pen_classes = 3
z_pen_logits = torch.randn(batch_size, num_pen_classes)

# x1_data, x2_data: coordinate data, shape (batch_size, num_components)
x1_data = torch.randn(batch_size, num_components)
x2_data = torch.randn(batch_size, num_components)

# pen_data: pen state data, shape (batch_size, 3)
# Last column will be used as pen_data[:, 2] for the fs calculation
pen_data = torch.zeros(batch_size, 3)
pen_data[:, 0] = torch.tensor([1.0, 0.0])  # class 0 for first sample, class 1 for second
pen_data[:, 1] = torch.tensor([0.0, 1.0])
pen_data[:, 2] = torch.tensor([0.1, 0.2])  # pen state values

ARGS = {
    'z_pi': z_pi,
    'z_mu1': z_mu1,
    'z_mu2': z_mu2,
    'z_sigma1': z_sigma1,
    'z_sigma2': z_sigma2,
    'z_corr': z_corr,
    'z_pen_logits': z_pen_logits,
    'x1_data': x1_data,
    'x2_data': x2_data,
    'pen_data': pen_data,
}
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 — 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
get_seq_from_gmm Ran
ran on a synthesized input
dailenson/SDT/models/gmm.py
code served (permissive licence) · get_code("0a53368641da7207")
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 Tensor, ndim 2, dtype family float, shape (2, 5) as a later run recorded it. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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 torch.nn as nn

torch.manual_seed(0)

# The function get_seq_from_gmm expects gmm_pred which will be reshaped to (-1, 123)
# So we need to create a tensor that when reshaped becomes (batch_size, 123)
# Let's make a batch of 2 samples, so total elements = 2 * 123 = 246

gmm_pred = torch.randn(246)

ARGS = {
    "gmm_pred": gmm_pred
}
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 Tensor, ndim 2, dtype family float, shape (2, 5) — shape (2, 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
image_resize Ran
ran on a synthesized input
koninik/htg_evaluation/utils/auxilary_functions.py
code served (permissive licence) · get_code("e8bf18bac8135a3c")
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 (8, 6, 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 small test image (4x3 RGB image)
img = np.random.rand(4, 3, 3).astype(np.float32)

# Define arguments for the function
ARGS = {
    'img': img,
    'height': 8,
    'width': None
}
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 (8, 6, 3) — shape (8, 6, 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
length_to_mask Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
dailenson/SDT/models/eval_model.py
code served (permissive licence) · get_code("61789a642c162a1b")
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 Tensor, ndim 2, dtype family float, shape (2, 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 torch

torch.manual_seed(0)

# Create a length tensor of shape [B] where B=2
# Values represent sequence lengths: [3, 5]
length = torch.tensor([3, 5], dtype=torch.long)

# Optional: max_len can be None (will use max of length) or an explicit int
max_len = None

# Optional: dtype can be None or a torch dtype for the output mask
dtype = torch.float32

ARGS = {
    "length": length,
    "max_len": max_len,
    "dtype": dtype,
}
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 Tensor, ndim 2, dtype family float, shape (2, 5) — shape (2, 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
mask_avr_pooling Ran
ran on a synthesized input
dailenson/SDT/models/eval_model.py
code served (permissive licence) · get_code("977232b5fb66d03c")
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 Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# x: (N, C, T) - batch size N=2, channels C=4, time steps T=3
x = torch.randn(2, 4, 3)

# l: (N,) - length tensor with batch size N=2, values should be <= T
l = torch.tensor([2, 3], dtype=torch.long)

ARGS = {
    "x": x,
    "l": l,
}
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 Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
mask_avr_pooling_rnn Ran
ran on a synthesized input
dailenson/SDT/models/eval_model.py
code served (permissive licence) · get_code("8acffa1b60bdf3a3")
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 Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# x: (N, T, C) tensor - batch of sequences with features
# l: (N,) tensor - lengths of each sequence in the batch

N = 2  # batch size
T = 5  # max sequence length
C = 4  # feature dimension

x = torch.randn(N, T, C)
l = torch.tensor([3, 5], dtype=torch.long)  # lengths for each sample in batch

ARGS = {
    "x": x,
    "l": l,
}
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 Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
performance Ran
ran on a synthesized input
koninik/htg_evaluation/htg_style.py
code served (permissive licence) · get_code("67e4aa07555e7628")
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 Tensor, 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 torch
import torch.nn as nn

torch.manual_seed(0)

# pred: output from model, should be logits of shape (batch_size, num_classes)
# CrossEntropyLoss expects pred to be (N, C) where N is batch size, C is number of classes
pred = torch.randn(2, 4)  # batch_size=2, num_classes=4

# label: target class indices of shape (batch_size,)
# CrossEntropyLoss expects label to be (N,) with values in range [0, C-1]
label = torch.tensor([0, 2])  # batch_size=2, values in range [0, 3]

ARGS = {
    "pred": pred,
    "label": label,
}
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 Tensor, 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
set_log Ran
ran on a synthesized input
dailenson/SDT/utils/logger.py
code served (permissive licence) · get_code("9f7efffab3bafdef")
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 dict. 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 time
import os
import tempfile

# Create a temporary directory for testing
temp_output_dir = tempfile.mkdtemp()

# Define sample arguments
output_dir = temp_output_dir
cfg_file = "/path/to/config.yaml"
log_name = "training_run"

ARGS = {
    "output_dir": output_dir,
    "cfg_file": cfg_file,
    "log_name": log_name,
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
tf_2d_normal Ran
ran on a synthesized input
dailenson/SDT/models/loss.py
code served (permissive licence) · get_code("67b19b6b4b262e2b")
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 Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Create small tensor arguments for the 2D normal distribution function
# All parameters should be tensors of the same shape for broadcasting
batch_size = 2
feature_dim = 4

x1 = torch.randn(batch_size, feature_dim)
x2 = torch.randn(batch_size, feature_dim)
mu1 = torch.randn(batch_size, feature_dim)
mu2 = torch.randn(batch_size, feature_dim)
s1 = torch.abs(torch.randn(batch_size, feature_dim)) + 0.1  # Ensure positive
s2 = torch.abs(torch.randn(batch_size, feature_dim)) + 0.1  # Ensure positive
rho = torch.tanh(torch.randn(batch_size, feature_dim))  # Ensure in (-1, 1)

ARGS = {
    "x1": x1,
    "x2": x2,
    "mu1": mu1,
    "mu2": mu2,
    "s1": s1,
    "s2": s2,
    "rho": rho,
}
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 Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
train_class_epoch Ran
ran on a synthesized input
koninik/htg_evaluation/htg_style.py
code served (permissive licence) · get_code("f0595038a8dda9b6")
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)

# Build a minimal model that takes an image and outputs logits
class MinimalModel(nn.Module):
    def __init__(self, num_classes=5):
        super().__init__()
        self.linear = nn.Linear(4, num_classes)
    
    def forward(self, x):
        # x shape: (batch, 4)
        return self.linear(x)

# Create a minimal args object with required attributes
class Args:
    def __init__(self):
        self.dataset = 'cifar'  # not 'iam'
        self.device = 'cpu'

# Build training data as a list of batches
# Each batch is a tuple where:
# - data[0] is the image (batch_size, 4)
# - data[1] is the label (batch_size,) for non-iam datasets
# - data[2] is the label (batch_size,) for iam datasets
training_data = []
for _ in range(3):  # 3 batches
    images = torch.randn(2, 4)  # batch_size=2, feature_dim=4
    labels = torch.randint(0, 5, (2,))  # 5 classes
    training_data.append((images, labels))

# Create model, optimizer, and args
model = MinimalModel(num_classes=5)
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)
args = Args()

ARGS = {
    'model': model,
    'training_data': training_data,
    'optimizer': optimizer,
    'args': args
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, tqdm — 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
assert_and_infer_cfg Not yet run
unverified (harvested, no recorded run)
dailenson/SDT/parse_config.py
code served (permissive licence) · get_code("87f89dec0143219f")
How it ran

no invocation recorded.

build_transformer Not yet run
unverified (harvested, no recorded run)
dailenson/SDT/models/transformer.py
code served (permissive licence) · get_code("d3c210804250a1dc")
How it ran

no invocation recorded.

Also present in other papers' code

Of 16 samples lifted from this paper's linked repositories, 16 carry a function identity and 0 do not. Of the 16 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 14 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
build_transformer 32 other papers
15 dated earlier, 17 later
2005.12872 (2020-05)
End-to-End Object Detection with Transformers
2103.12115 (2021-03)
End-to-End Trainable Multi-Instance Pose Estimation with Transformers
2104.04369 (2021-04)
Video-aided Unsupervised Grammar Induction
and 29 more not named here
length_to_mask 1 other paper
1 later
2307.12058 (2023-07)
Discovering Spatio-Temporal Rationales for Video Question Answering

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

Training machines to synthesize diverse handwritings is an intriguing task. Recently, RNN-based methods have been proposed to generate stylized online Chinese characters. However, these methods mainly focus on capturing a person's overall writing style, neglecting subtle style inconsistencies between characters written by the same person. For example, while a person's handwriting typically exhibits general uniformity (e.g., glyph slant and aspect ratios), there are still small style variations in finer details (e.g., stroke length and curvature) of characters. In light of this, we propose to disentangle the style representations at both writer and character levels from individual handwritings to synthesize realistic stylized online handwritten characters. Specifically, we present the styledisentangled Transformer (SDT), which employs two complementary contrastive objectives to extract the style commonalities of reference samples and capture the detailed style patterns of each sample, respectively. Extensive experiments on various language scripts demonstrate the effectiveness of SDT. Notably, our empirical findings reveal that the two learned style representations provide information at different frequency magnitudes, underscoring the importance of separate style extraction. Our source code is public at: https://github.com/dailenson/SDT.

For agents

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

get_harvested_code_for_paper("2303.14736")
get_code_for_paper("2303.14736")
have("2303.14736")

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

curl https://syntology.ai/api/ran/2303.14736.json

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

[![Syntology run record](https://syntology.ai/api/ran/2303.14736.svg)](https://syntology.ai/paper/2303.14736)

Connect an agent — have() is free.