SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2407.11910 · NeurIPS · 2024

Benchmarking the Attribution Quality of Vision Models

Stefan Roth, Simone Schaub-Meyer, Robin Hesse

arXiv · PDF · Open in the Atlas

Code that ran

5 of 14 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; 5 failures of our instrument; 9 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/2407.11910.json · badge: /api/ran/2407.11910.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
visinf/idsds canonical (classifier label)
classifier basis: census
5 ran · 9 no recorded run
FunctionStatusWhere it lives
compute_rollout_attention Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
visinf/idsds/models/ViT/ViT_LRP.py
code served (permissive licence) · get_code("927e684e71ca3ee9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned Tensor, ndim 3, dtype family float, shape (2, 5, 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)

batch_size = 2
num_tokens = 5
num_layers = 3

# Create a list of attention matrices, one per layer
# Each matrix is of shape (batch_size, num_tokens, num_tokens)
all_layer_matrices = []
for i in range(num_layers):
    # Create random attention-like matrices (should sum to 1 along last dim for realism, but not required)
    attn_matrix = torch.rand(batch_size, num_tokens, num_tokens)
    attn_matrix = attn_matrix / attn_matrix.sum(dim=-1, keepdim=True)
    all_layer_matrices.append(attn_matrix)

start_layer = 1

ARGS = {
    'all_layer_matrices': all_layer_matrices,
    'start_layer': start_layer
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
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 3, dtype family float, shape (2, 5, 5) — shape (2, 5, 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
compute_rollout_attention Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
visinf/idsds/models/ViT/ViT_explanation_generator.py
code served (permissive licence) · get_code("5a98eb7226d25564")
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 3, dtype family float, shape (2, 4, 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)

# The function expects:
# - all_layer_matrices: a list of attention matrices
# - start_layer: an integer index

# Each attention matrix should be shape (batch_size, num_tokens, num_tokens)
# Let's create a small example with batch_size=2, num_tokens=4

batch_size = 2
num_tokens = 4
num_layers = 3

# Create attention matrices for multiple layers
all_layer_matrices = [
    torch.rand(batch_size, num_tokens, num_tokens) for _ in range(num_layers)
]

# Normalize them to be valid attention matrices (sum to 1 along last dimension)
for i in range(len(all_layer_matrices)):
    all_layer_matrices[i] = all_layer_matrices[i] / all_layer_matrices[i].sum(dim=-1, keepdim=True)

ARGS = {
    "all_layer_matrices": all_layer_matrices,
    "start_layer": 0
}
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 3, dtype family float, shape (2, 4, 4) — shape (2, 4, 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
conv1x1 Ran
ran; our contract draft was wrong, not the code
visinf/idsds/models/resnet.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from medaitech/scrd4ad, is pointer only (licence: NONE) · get_code("2a80220dabcb742a")
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 medaitech/scrd4ad, 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.

conv3x3 Ran
ran; our contract draft was wrong, not the code
visinf/idsds/models/resnet.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from medaitech/scrd4ad, is pointer only (licence: NONE) · get_code("600ff2c45e0de056")
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 medaitech/scrd4ad, 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.

make_layers Ran
ran; our contract draft was wrong, not the code
visinf/idsds/models/vgg.py
code served (permissive licence) · get_code("5adf613fdebfcbd4")
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)

# cfg is a list of Union[str, int] representing layer configuration
# It can contain integers (number of output channels) and 'M' (MaxPool2d)
cfg = [64, 'M', 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512]

# batch_norm is a boolean flag
batch_norm = False

ARGS = {
    'cfg': cfg,
    'batch_norm': batch_norm
}
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
resnet18 Not yet run
unverified (harvested, no recorded run)
visinf/idsds/models/resnet.py
code served (permissive licence) · get_code("8c0113732a399cea")
How it ran

no invocation recorded.

vgg11 Not yet run
unverified (harvested, no recorded run)
visinf/idsds/models/vgg.py
code served (permissive licence) · get_code("76b1bf0c7e996f90")
How it ran

no invocation recorded.

vgg11_bn Not yet run
unverified (harvested, no recorded run)
visinf/idsds/models/vgg.py
code served (permissive licence) · get_code("df6cfaa1fbccd91a")
How it ran

no invocation recorded.

vit_base_patch16_224 Not yet run
unverified (harvested, no recorded run)
visinf/idsds/models/ViT/ViT_LRP.py
code served (permissive licence) · get_code("5fe2124317452946")
How it ran

no invocation recorded.

vit_base_patch16_224 Not yet run
unverified (harvested, no recorded run)
visinf/idsds/models/ViT/ViT_gradient_attribution.py
code served (permissive licence) · get_code("5406cd8a829fda92")
How it ran

no invocation recorded.

vit_base_patch16_224 Not yet run
unverified (harvested, no recorded run)
visinf/idsds/models/ViT/ViT_new.py
code served (permissive licence) · get_code("57e03ac784886a52")
How it ran

no invocation recorded.

vit_large_patch16_224 Not yet run
unverified (harvested, no recorded run)
visinf/idsds/models/ViT/ViT_LRP.py
code served (permissive licence) · get_code("005867bd71baa4bb")
How it ran

no invocation recorded.

vit_large_patch16_224 Not yet run
unverified (harvested, no recorded run)
visinf/idsds/models/ViT/ViT_gradient_attribution.py
code served (permissive licence) · get_code("1b096e0f5504ad57")
How it ran

no invocation recorded.

vit_large_patch16_224 Not yet run
unverified (harvested, no recorded run)
visinf/idsds/models/ViT/ViT_new.py
code served (permissive licence) · get_code("fd65fde98f9bb5fc")
How it ran

no invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 12 distinct functions with an identity, 12 are also present in other papers whose code we hold (12 with at least one paper dated earlier, 3 the same month, 5 later) and 0 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
conv1x1 112 other papers
84 dated earlier, 2 the same month, 22 later, 4 same year, order not determinable · 6 only as an assembly brick
1512.03385 (2015-12)
Deep Residual Learning for Image Recognition
1806.01531 (2018-06)
Deep Mixture of Experts via Shallow Embedding
1905.12794 (2019-05)
Fashion IQ: A New Dataset Towards Retrieving Images by Natural Language Feedback
and 109 more not named here
conv3x3 111 other papers
84 dated earlier, 2 the same month, 21 later, 4 same year, order not determinable · 6 only as an assembly brick
1512.03385 (2015-12)
Deep Residual Learning for Image Recognition
1806.01531 (2018-06)
Deep Mixture of Experts via Shallow Embedding
1905.12794 (2019-05)
Fashion IQ: A New Dataset Towards Retrieving Images by Natural Language Feedback
and 108 more not named here
resnet18 36 other papers
31 dated earlier, 1 the same month, 3 later, 1 same year, order not determinable
1905.12794 (2019-05)
Fashion IQ: A New Dataset Towards Retrieving Images by Natural Language Feedback
2002.09564 (2020-02)
Towards Robust and Reproducible Active Learning Using Neural Networks
2003.04297 (2020-03)
Improved Baselines with Momentum Contrastive Learning
and 33 more not named here
make_layers 9 other papers
7 dated earlier, 2 later
2002.09564 (2020-02)
Towards Robust and Reproducible Active Learning Using Neural Networks
2306.12685 (2023-06)
Rethinking the Backward Propagation for Adversarial Transferability
2306.13856 (2023-06)
Learning-to-Rank Meets Language: Boosting Language-Driven Ordering Alignment for Ordinal Classification
and 6 more not named here
compute_rollout_attention 4 other papers
3 dated earlier, 1 later
2205.09671 (2022-05)
A graph-transformer for whole slide image classification
2206.10552 (2022-06)
Vicinity Vision Transformer
2308.06248 (2023-08)
FunnyBirds: A Synthetic Vision Dataset for a Part-Based Analysis of Explainable AI Methods
and 1 more not named here
vgg11 3 other papers
3 dated earlier
2306.13856 (2023-06)
Learning-to-Rank Meets Language: Boosting Language-Driven Ordering Alignment for Ordinal Classification
2308.06248 (2023-08)
FunnyBirds: A Synthetic Vision Dataset for a Part-Based Analysis of Explainable AI Methods
2312.06372 (2023-12)
Ternary Spike: Learning Ternary Spikes for Spiking Neural Networks
vgg11_bn 3 other papers
3 dated earlier
2306.13856 (2023-06)
Learning-to-Rank Meets Language: Boosting Language-Driven Ordering Alignment for Ordinal Classification
2308.06248 (2023-08)
FunnyBirds: A Synthetic Vision Dataset for a Part-Based Analysis of Explainable AI Methods
2312.06372 (2023-12)
Ternary Spike: Learning Ternary Spikes for Spiking Neural Networks
compute_rollout_attention 2 other papers
2 dated earlier
2308.06248 (2023-08)
FunnyBirds: A Synthetic Vision Dataset for a Part-Based Analysis of Explainable AI Methods
2403.01590 (2024-03)
The Hidden Attention of Mamba Models
vit_base_patch16_224 2 other papers
2 dated earlier
2110.07719 (2021-10)
Certified Patch Robustness via Smoothed Vision Transformers
2308.06248 (2023-08)
FunnyBirds: A Synthetic Vision Dataset for a Part-Based Analysis of Explainable AI Methods
vit_base_patch16_224 1 other paper
1 dated earlier
2308.06248 (2023-08)
FunnyBirds: A Synthetic Vision Dataset for a Part-Based Analysis of Explainable AI Methods
vit_large_patch16_224 1 other paper
1 dated earlier
2308.06248 (2023-08)
FunnyBirds: A Synthetic Vision Dataset for a Part-Based Analysis of Explainable AI Methods
vit_large_patch16_224 1 other paper
1 dated earlier
2308.06248 (2023-08)
FunnyBirds: A Synthetic Vision Dataset for a Part-Based Analysis of Explainable AI Methods

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

Attribution maps are one of the most established tools to explain the functioning of computer vision models. They assign importance scores to input features, indicating how relevant each feature is for the prediction of a deep neural network. While much research has gone into proposing new attribution methods, their proper evaluation remains a difficult challenge. In this work, we propose a novel evaluation protocol that overcomes two fundamental limitations of the widely used incremental-deletion protocol, i.e., the out-of-domain issue and lacking inter-model comparisons. This allows us to evaluate 23 attribution methods and how different design choices of popular vision backbones affect their attribution quality. We find that intrinsically explainable models outperform standard models and that raw attribution values exhibit a higher attribution quality than what is known from previous work. Further, we show consistent changes in the attribution quality when varying the network design, indicating that some standard design choices promote attribution quality. 1

For agents

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

get_harvested_code_for_paper("2407.11910")
get_code_for_paper("2407.11910")
have("2407.11910")

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

curl https://syntology.ai/api/ran/2407.11910.json

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

[![Syntology run record](https://syntology.ai/api/ran/2407.11910.svg)](https://syntology.ai/paper/2407.11910)

Connect an agent — have() is free.