SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2401.08501 · ICLR · 2024

ValUES: A Framework for Systematic Validation of Uncertainty Estimation in Semantic Segmentation

Klaus Maier-Hein, Maximilian Zenk, Paul Jaeger, Kim-Celine Kahl, Carsten Üth

arXiv · PDF · Open in the Atlas

Code that ran

7 of 10 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 0 have no contract verdict; 7 failures of our instrument; 3 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/2401.08501.json · badge: /api/ran/2401.08501.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
IML-DKFZ/values canonical (classifier label)
classifier basis: census
7 ran · 1 no recorded run
iml-dkfz/values canonical (classifier label)
classifier basis: classifier_2025
0 ran · 2 no recorded run
FunctionStatusWhere it lives
conv3x3 Ran
ran; our contract draft was wrong, not the code
IML-DKFZ/values/uncertainty_modeling/models/hrnet_module.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from lixiaotong97/DSU, is pointer only (licence: NONE) · get_code("fac5364e2f53c6db")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from lixiaotong97/DSU, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_train_data_samples Ran
ran; our contract draft was wrong, not the code
IML-DKFZ/values/uncertainty_modeling/lidc_idri_datamodule_3D.py
code served (permissive licence) · get_code("e5c66f25e43363f2")
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 os
import tempfile
import numpy as np

# Create a temporary directory structure with sample .npy files
temp_dir = tempfile.mkdtemp()
images_dir = os.path.join(temp_dir, 'images')
labels_dir = os.path.join(temp_dir, 'labels')
os.makedirs(images_dir)
os.makedirs(labels_dir)

# Create sample image files
image_filenames = ['sample_001.npy', 'sample_002.npy', 'sample_003.npy']
for img_file in image_filenames:
    np.save(os.path.join(images_dir, img_file), np.random.rand(10, 10, 10))

# Create sample label files
label_filenames = [
    'sample_001_00_mask.npy',
    'sample_001_01_mask.npy',
    'sample_002_00_mask.npy',
    'sample_003_00_mask.npy',
    'sample_003_01_mask.npy',
]
for label_file in label_filenames:
    np.save(os.path.join(labels_dir, label_file), np.random.randint(0, 2, (10, 10, 10)))

ARGS = {
    'base_dir': temp_dir,
    'pattern': '*.npy',
    'subject_ids': ['sample_001.npy', 'sample_002.npy'],
    'num_raters': 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 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_train_data_samples Ran
ran; our contract draft was wrong, not the code
IML-DKFZ/values/uncertainty_modeling/toy_datamodule_3D.py
code served (permissive licence) · get_code("efa971e5017cd37b")
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 os
import tempfile
import numpy as np

# Create a temporary directory structure with sample data
temp_dir = tempfile.mkdtemp()
images_dir = os.path.join(temp_dir, 'imagesTr')
labels_dir = os.path.join(temp_dir, 'labelsTr')
os.makedirs(images_dir)
os.makedirs(labels_dir)

# Create sample .npy files
sample_image_data = np.random.rand(32, 32, 32).astype(np.float32)
sample_label_data = np.random.randint(0, 2, (32, 32, 32)).astype(np.uint8)

# Create image files
np.save(os.path.join(images_dir, 'subject_001.npy'), sample_image_data)
np.save(os.path.join(images_dir, 'subject_002.npy'), sample_image_data)

# Create label files for multiple raters
np.save(os.path.join(labels_dir, 'subject_001_00.npy'), sample_label_data)
np.save(os.path.join(labels_dir, 'subject_001_01.npy'), sample_label_data)
np.save(os.path.join(labels_dir, 'subject_002_00.npy'), sample_label_data)
np.save(os.path.join(labels_dir, 'subject_002_01.npy'), sample_label_data)

ARGS = {
    'base_dir': temp_dir,
    'pattern': '*.npy',
    'subject_ids': None,
    'num_raters': 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 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_val_test_data_samples Ran
ran; our contract draft was wrong, not the code
IML-DKFZ/values/uncertainty_modeling/lidc_idri_datamodule_3D.py
code served (permissive licence) · get_code("b10eee998c14b87a")
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 os
import tempfile
import numpy as np

# Create a temporary directory structure with mock data
tmpdir = tempfile.mkdtemp()
images_dir = os.path.join(tmpdir, 'images')
labels_dir = os.path.join(tmpdir, 'labels')
os.makedirs(images_dir)
os.makedirs(labels_dir)

# Create mock image and label files
image_data = np.random.rand(128, 128, 128).astype(np.float32)
np.save(os.path.join(images_dir, 'subject_001.npy'), image_data)

label_data = np.random.randint(0, 2, (128, 128, 128)).astype(np.uint8)
np.save(os.path.join(labels_dir, 'subject_001_00_mask.npy'), label_data)
np.save(os.path.join(labels_dir, 'subject_001_01_mask.npy'), label_data)

ARGS = {
    'base_dir': tmpdir,
    'pattern': '*.npy',
    'subject_ids': ['subject_001.npy'],
    'num_raters': 2,
    'test': False,
    'patch_size': 64,
    'patch_overlap': 1.0
}
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 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_val_test_data_samples Ran
ran; our contract draft was wrong, not the code
IML-DKFZ/values/uncertainty_modeling/toy_datamodule_3D.py
code served (permissive licence) · get_code("ce47586ffdff2741")
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 tempfile
import os
import numpy as np

# Create a temporary directory structure for the test
temp_dir = tempfile.mkdtemp()
images_tr_dir = os.path.join(temp_dir, 'imagesTr')
labels_tr_dir = os.path.join(temp_dir, 'labelsTr')
images_ts_dir = os.path.join(temp_dir, 'imagesTs')
labels_ts_dir = os.path.join(temp_dir, 'labelsTs')

os.makedirs(images_tr_dir, exist_ok=True)
os.makedirs(labels_tr_dir, exist_ok=True)
os.makedirs(images_ts_dir, exist_ok=True)
os.makedirs(labels_ts_dir, exist_ok=True)

# Create sample numpy arrays for training data
train_image = np.random.rand(128, 128, 128).astype(np.float32)
train_label_0 = np.random.randint(0, 2, (128, 128, 128)).astype(np.int32)
train_label_1 = np.random.randint(0, 2, (128, 128, 128)).astype(np.int32)

np.save(os.path.join(images_tr_dir, 'subject_001.npy'), train_image)
np.save(os.path.join(labels_tr_dir, 'subject_001_00.npy'), train_label_0)
np.save(os.path.join(labels_tr_dir, 'subject_001_01.npy'), train_label_1)

# Create sample numpy arrays for test data
test_image = np.random.rand(128, 128, 128).astype(np.float32)
test_label_0 = np.random.randint(0, 2, (128, 128, 128)).astype(np.int32)

np.save(os.path.join(images_ts_dir, 'subject_002.npy'), test_image)
np.save(os.path.join(labels_ts_dir, 'subject_002_00.npy'), test_label_0)

ARGS = {
    'base_dir': temp_dir,
    'pattern': '*.npy',
    'subject_ids': None,
    'num_raters': 2,
    'test': False,
    'patch_size': 64,
    'patch_overlap': 1.0
}
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 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
soft_dice Ran
ran; our contract draft was wrong, not the code
IML-DKFZ/values/uncertainty_modeling/loss_modules.py
code served (permissive licence) · get_code("5775655dbdd56e51")
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
from torch import nn, Tensor

torch.manual_seed(0)

# Construct valid arguments for soft_dice function
# The function expects:
# - net_output: predicted probabilities (Tensor)
# - gt: ground truth one-hot encoded (Tensor)
# - smooth: smoothing in denominator (float)
# - smooth_in_nom: smoothing in nominator (float)
# - only_intersect: whether to return only intersect or mean (bool)

# Create small tensors with batch size 2, 4 classes
# Shape: (batch, classes, spatial_dims...)
batch_size = 2
num_classes = 4
spatial_size = 3

# net_output: predicted probabilities (batch x classes x spatial)
net_output = torch.rand(batch_size, num_classes, spatial_size, requires_grad=True)
# Apply softmax to ensure valid probabilities
net_output = torch.softmax(net_output, dim=1)

# gt: ground truth one-hot encoded (batch x classes x spatial)
# Create one-hot by selecting random class per spatial location
gt_indices = torch.randint(0, num_classes, (batch_size, spatial_size))
gt = torch.zeros(batch_size, num_classes, spatial_size)
for b in range(batch_size):
    for s in range(spatial_size):
        gt[b, gt_indices[b, s], s] = 1.0

# Define other parameters
smooth = 1.0
smooth_in_nom = 1.0
only_intersect = False

ARGS = {
    "net_output": net_output,
    "gt": gt,
    "smooth": smooth,
    "smooth_in_nom": smooth_in_nom,
    "only_intersect": only_intersect
}
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
subfiles Ran
ran; our contract draft was wrong, not the code
IML-DKFZ/values/uncertainty_modeling/lidc_idri_datamodule_3D.py
code served (permissive licence) · get_code("03cb8b894670bbd8")
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 os
import tempfile
import shutil

# Create a temporary directory with some test files
temp_dir = tempfile.mkdtemp()

# Create some test files with various prefixes and suffixes
test_files = [
    "model_v1.txt",
    "model_v2.txt",
    "data_train.csv",
    "data_test.csv",
    "config.json",
    "readme.md"
]

for filename in test_files:
    filepath = os.path.join(temp_dir, filename)
    with open(filepath, 'w') as f:
        f.write("test content")

# Define arguments for the subfiles function
ARGS = {
    "folder": temp_dir,
    "join": True,
    "prefix": None,
    "suffix": None,
    "sort": True
}
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
SsnUNet3D Not yet run
unverified (harvested, no recorded run)
iml-dkfz/values/uncertainty_modeling/models/ssn_unet3D_module.py
code served (permissive licence) · get_code("42ad3b72fe54e6f3")
How it ran

no invocation recorded.

UNet3D Not yet run
unverified (harvested, no recorded run)
iml-dkfz/values/uncertainty_modeling/models/ssn_unet3D_module.py
code served (permissive licence) · get_code("48c5ca3e2ad0db57")
How it ran

no invocation recorded.

get_seg_model Not yet run
unverified (harvested, no recorded run)
IML-DKFZ/values/uncertainty_modeling/models/hrnet_module.py
code served (permissive licence) · get_code("f47754874c0be188")
How it ran

no invocation recorded.

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 10 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 1 the same month, 1 later) and 9 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
conv3x3 686 other papers
548 dated earlier, 4 the same month, 117 later, 17 same year, order not determinable · 31 only as an assembly brick
1406.2661 (2014-06)
Generative Adversarial Networks
1409.0575 (2014-09)
ImageNet Large Scale Visual Recognition Challenge
1412.6572 (2014-12)
Explaining and Harnessing Adversarial Examples
and 683 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

Uncertainty estimation is an essential and heavily-studied component for the reliable application of semantic segmentation methods. While various studies exist claiming methodological advances on the one hand, and successful application on the other hand, the field is currently hampered by a gap between theory and practice leaving fundamental questions unanswered: Can data-related and model-related uncertainty really be separated in practice? Which components of an uncertainty method are essential for real-world performance? Which uncertainty method works well for which application? In this work, we link this research gap to a lack of systematic and comprehensive evaluation of uncertainty methods. Specifically, we identify three key pitfalls in current literature and present an evaluation framework that bridges the research gap by providing 1) a controlled environment for studying data ambiguities as well as distribution shifts, 2) systematic ablations of relevant method components, and 3) test-beds for the five predominant uncertainty applications: OoD-detection, active learning, failure detection, calibration, and ambiguity modeling. Empirical results on simulated as well as real-world data demonstrate how the proposed framework is able to answer the predominant questions in the field revealing for instance that 1) separation of uncertainty types works on simulated data but does not necessarily translate to real-world data, 2) aggregation of scores is a crucial but currently neglected component of uncertainty methods, 3) While ensembles are performing most robustly across the different downstream tasks and settings, test-time augmentation often constitutes a light-weight alternative. Code is at: https://github.com/IML-DKFZ/values

For agents

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

get_harvested_code_for_paper("2401.08501")
get_code_for_paper("2401.08501")
have("2401.08501")

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

curl https://syntology.ai/api/ran/2401.08501.json

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

[![Syntology run record](https://syntology.ai/api/ran/2401.08501.svg)](https://syntology.ai/paper/2401.08501)

Connect an agent — have() is free.