SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2305.08455 · ICCV · 2023

Document Understanding Dataset and Evaluation (DUDE )

K Leuven, Matthew Blaschko, Łukasz Borchmann, Dawid Jurkiewicz, Michał Pietruszka, Mickaël Coustaty, Paweł Józiak, Jordy Landeghem, Rubèn Tito, Rafał Powalski, Bertrand Ackaert, Ernest Valveny, and 4 more

arXiv · PDF · Open in the Atlas

Code that ran

4 of 5 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 4 have no contract verdict; 0 failures of our instrument; 1 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/2305.08455.json · badge: /api/ran/2305.08455.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
rubenpt91/MP-DocVQA-Framework canonical (classifier label)
classifier basis: census
4 ran · 1 no recorded run
FunctionStatusWhere it lives
build_dataset Ran
ran on a synthesized input
rubenpt91/MP-DocVQA-Framework/build_utils.py
code served (permissive licence) · get_code("88a281f34cff4e96")
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 MockSPDocVQA. 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 unittest.mock import Mock, MagicMock

torch.manual_seed(0)

# Create mock dataset classes that satisfy the function's requirements
class MockSPDocVQA:
    def __init__(self, imdb_dir, images_dir, split, dataset_kwargs):
        self.imdb_dir = imdb_dir
        self.images_dir = images_dir
        self.split = split
        self.dataset_kwargs = dataset_kwargs

class MockMPDocVQA:
    def __init__(self, imdb_dir, images_dir, page_retrieval, split, dataset_kwargs):
        self.imdb_dir = imdb_dir
        self.images_dir = images_dir
        self.page_retrieval = page_retrieval
        self.split = split
        self.dataset_kwargs = dataset_kwargs

class MockDUDE:
    def __init__(self, imdb_dir, images_dir, page_retrieval, split, dataset_kwargs):
        self.imdb_dir = imdb_dir
        self.images_dir = images_dir
        self.page_retrieval = page_retrieval
        self.split = split
        self.dataset_kwargs = dataset_kwargs

# Mock the imports at module level
import sys
from types import ModuleType

datasets_module = ModuleType('datasets')
sp_docvqa_module = ModuleType('SP_DocVQA')
mp_docvqa_module = ModuleType('MP_DocVQA')
dude_module = ModuleType('DUDE')

sp_docvqa_module.SPDocVQA = MockSPDocVQA
mp_docvqa_module.MPDocVQA = MockMPDocVQA
dude_module.DUDE = MockDUDE

datasets_module.SP_DocVQA = sp_docvqa_module
datasets_module.MP_DocVQA = mp_docvqa_module
datasets_module.DUDE = dude_module

sys.modules['datasets'] = datasets_module
sys.modules['datasets.SP_DocVQA'] = sp_docvqa_module
sys.modules['datasets.MP_DocVQA'] = mp_docvqa_module
sys.modules['datasets.DUDE'] = dude_module

# Define configuration dict
config = {
    'model_name': 'layoutlmv2',
    'dataset_name': 'SP-DocVQA',
    'imdb_dir': '/path/to/imdb',
    'images_dir': '/path/to/images',
    'page_retrieval': False,
    'max_pages': 1
}

# Define split parameter
split = 'train'

ARGS = {
    'config': config,
    'split': split
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
datasets, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MockSPDocVQA — a MockSPDocVQA output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a MockSPDocVQA 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
build_model Ran
ran on a synthesized input
rubenpt91/MP-DocVQA-Framework/build_utils.py
code served (permissive licence) · get_code("2d363ded709cf8c9")
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 MockBertQA. 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
import torch.nn as nn

torch.manual_seed(0)

# Create a minimal mock model class that satisfies the function's usage
class MockModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.model = nn.Linear(4, 3)
    
    def parallelize(self):
        pass

class MockBertQA(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.model = nn.Linear(4, 3)
    
    def parallelize(self):
        pass

class MockLongformer(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.model = nn.Linear(4, 3)
    
    def parallelize(self):
        pass

class MockBigBird(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.model = nn.Linear(4, 3)
    
    def parallelize(self):
        pass

class MockLayoutLMv2(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.model = nn.Linear(4, 3)
    
    def parallelize(self):
        pass

class MockLayoutLMv3(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.model = nn.Linear(4, 3)
    
    def parallelize(self):
        pass

class MockT5(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.model = nn.Linear(4, 3)
    
    def parallelize(self):
        pass

class MockVT5(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.model = nn.Linear(4, 3)
    
    def parallelize(self):
        pass

class MockHiVT5(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.model = nn.Linear(4, 3)
    
    def parallelize(self):
        pass

# Patch the imports to return our mocks
import sys
from unittest.mock import MagicMock

models_module = MagicMock()
models_module.BertQA.BertQA = MockBertQA
models_module.Longformer.Longformer = MockLongformer
models_module.BigBird.BigBird = MockBigBird
models_module.LayoutLMv2.LayoutLMv2 = MockLayoutLMv2
models_module.LayoutLMv3.LayoutLMv3 = MockLayoutLMv3
models_module.T5.T5 = MockT5
models_module.VT5.ProxyVT5 = MockVT5
models_module.HiVT5.Proxy_HiVT5 = MockHiVT5

sys.modules['models'] = MagicMock()
sys.modules['models.BertQA'] = MagicMock(BertQA=MockBertQA)
sys.modules['models.Longformer'] = MagicMock(Longformer=MockLongformer)
sys.modules['models.BigBird'] = MagicMock(BigBird=MockBigBird)
sys.modules['models.LayoutLMv2'] = MagicMock(LayoutLMv2=MockLayoutLMv2)
sys.modules['models.LayoutLMv3'] = MagicMock(LayoutLMv3=MockLayoutLMv3)
sys.modules['models.T5'] = MagicMock(T5=MockT5)
sys.modules['models.VT5'] = MagicMock(ProxyVT5=MockVT5)
sys.modules['models.HiVT5'] = MagicMock(Proxy_HiVT5=MockHiVT5)

config = {
    'model_name': 'bertqa',
    'device': 'cpu',
    'data_parallel': False
}

ARGS = {
    'config': config
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
models, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MockBertQA — a MockBertQA output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a MockBertQA 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
mpdocvqa_collate_fn Ran
ran on a synthesized input
rubenpt91/MP-DocVQA-Framework/datasets/MP_DocVQA.py
code served (permissive licence) · get_code("74216735ae11b6fd")
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 torch
import numpy as np

torch.manual_seed(0)

# mpdocvqa_collate_fn expects a batch parameter
# which is a list of dictionaries
# The function reconstructs it so each key maps to a list of values

# Create sample batch items - each is a dictionary
batch_item_1 = {
    "image": torch.randn(4, 3),
    "question": "What is this?",
    "answer": "text answer",
    "page_id": 1
}

batch_item_2 = {
    "image": torch.randn(4, 3),
    "question": "Where is this?",
    "answer": "location answer",
    "page_id": 2
}

batch = [batch_item_1, batch_item_2]

ARGS = {
    "batch": batch
}
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 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
singlepage_docvqa_collate_fn Ran
ran on a synthesized input
rubenpt91/MP-DocVQA-Framework/datasets/SP_DocVQA.py
code served (permissive licence) · get_code("d7daacc3e799e5e2")
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 torch
import numpy as np

torch.manual_seed(0)

# The function expects a batch which is a list of dictionaries
# It extracts values by key from each dictionary in the batch
# and reorganizes them into a single dict with lists as values

# Create sample batch data - list of dictionaries
sample_dict_1 = {
    "image": torch.randn(2, 3, 4, 4),  # small image tensor
    "question": torch.tensor([1, 2, 3, 4]),
    "answer": "sample answer 1",
    "metadata": {"page": 0}
}

sample_dict_2 = {
    "image": torch.randn(2, 3, 4, 4),  # same structure
    "question": torch.tensor([5, 6, 7, 8]),
    "answer": "sample answer 2",
    "metadata": {"page": 1}
}

batch = [sample_dict_1, sample_dict_2]

ARGS = {
    "batch": batch
}
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 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
build_optimizer Not yet run
unverified (harvested, no recorded run)
rubenpt91/MP-DocVQA-Framework/build_utils.py
code served (permissive licence) · get_code("d7bf1e31e1918784")
How it ran

no invocation recorded.

Also present in other papers' code

Of 5 samples lifted from this paper's linked repositories, 5 carry a function identity and 0 do not. Of the 5 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 5 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.

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

We call on the Document AI (DocAI) community to reevaluate current methodologies and embrace the challenge of creating more practically-oriented benchmarks. Document Understanding Dataset and Evaluation (DUDE) seeks to remediate the halted research progress in understanding visually-rich documents (VRDs). We present a new dataset 1 with novelties related to types of questions, answers, and document layouts based on multi-industry, multi-domain, and multi-page VRDs of various origins, and dates. Moreover, we are pushing the boundaries of current methods by creating multi-task and multi-domain evaluation setups that more accurately simulate real-world situations where powerful generalization and adaptation under low-resource settings are desired. DUDE aims to set a new standard as a more practical, long-standing benchmark for the community, and we hope that it will lead to future extensions and contributions that address real-world challenges. Finally, our work illustrates the importance of finding more efficient ways to model language, images, and layout in DocAI.

For agents

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

get_harvested_code_for_paper("2305.08455")
get_code_for_paper("2305.08455")
have("2305.08455")

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

curl https://syntology.ai/api/ran/2305.08455.json

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

[![Syntology run record](https://syntology.ai/api/ran/2305.08455.svg)](https://syntology.ai/paper/2305.08455)

Connect an agent — have() is free.