SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2401.07803 · ACL · 2024

Uncovering the Full Potential of Visual Grounding Methods in VQA

Daniel Reich, Tanja Schultz

arXiv · PDF · Open in the Atlas

Code that ran

9 of 9 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; 2 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 7 failures of our instrument; 0 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/2401.07803.json · badge: /api/ran/2401.07803.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
dreichCSL/TrueVG canonical (classifier label)
classifier basis: census
9 ran · 0 no recorded run
FunctionStatusWhere it lives
compute_score_with_logits Ran
ran; our contract draft was wrong, not the code
dreichCSL/TrueVG/components/losses.py
code served (permissive licence) · get_code("89bdd6decba2b415")
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. 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 expects:
# 1. logits: a tensor that can be passed to torch.max(logits, 1) - needs shape (batch_size, num_classes)
# 2. labels: a tensor that has .size() method and supports scatter_ operations - needs same shape as logits for element-wise multiplication

batch_size = 2
num_classes = 4

# Create logits tensor with shape (batch_size, num_classes)
logits = torch.randn(batch_size, num_classes)

# Create labels tensor with shape (batch_size, num_classes)
# This should be one-hot encoded or binary labels for the scores computation to make sense
labels = torch.zeros(batch_size, num_classes)
labels[0, 1] = 1
labels[1, 3] = 1

ARGS = {
    "logits": logits,
    "labels": labels
}
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
find_infusion_objects Ran
ran; our contract draft was wrong, not the code
dreichCSL/TrueVG/preprocessing/truevg/prep_infusion.py
code served (permissive licence) · get_code("69f8cd8f0aee84d4")
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 pickle
import json
import h5py
import numpy as np
import tempfile
import os

# Create a temporary directory for test data
temp_dir = tempfile.mkdtemp()
preprocessing_dir = os.path.join(temp_dir, 'preprocessing')
os.makedirs(preprocessing_dir, exist_ok=True)

# Create minimal GQA reference JSON
reference_gqa_json = {
    'img_001': {'index': 0, 'objectsNum': 3},
    'img_002': {'index': 1, 'objectsNum': 2},
}
with open(os.path.join(temp_dir, 'output_gqa_detectron_objects_info.json'), 'w') as f:
    json.dump(reference_gqa_json, f)

# Create object name classes pickle
objectname_classes = {
    'dog': 0,
    'cat': 1,
    'person': 2,
    'car': 3,
    'chair': 4,
}
with open(os.path.join(preprocessing_dir, 'objectname_classes.pkl'), 'wb') as f:
    pickle.dump(objectname_classes, f)

# Create attribute categories pickle
attribute_categories = {
    'color': ['red', 'blue', 'green'],
    'size': ['small', 'large'],
}
with open(os.path.join(preprocessing_dir, 'attribute_categories.pkl'), 'wb') as f:
    pickle.dump(attribute_categories, f)

# Create HDF5 image database
with h5py.File(os.path.join(preprocessing_dir, 'img_db.h5'), 'w') as hf:
    # Create datasets for object names and attributes
    # img_index 0: 3 objects; img_index 1: 2 objects
    name_data = np.array([[0, 1, 2], [3, 4, -1]], dtype=np.int32)
    hf.create_dataset('name', data=name_data)
    
    # Attributes: -1 means no attributes, otherwise indices into attribute categories
    attr_data = np.array([
        [[0, -1], [1, 0], [0, 1]],  # img_index 0: 3 objects with attributes
        [[1, 1], [0, -1], [-1, -1]],  # img_index 1: 2 objects
    ], dtype=np.int32)
    hf.create_dataset('attributes', data=attr_data)

# Create scene graph (sg)
sg = {
    'img_001': {
        'objects': {
            0: {'name': 'dog', 'attributes': ['red', 'small']},
            1: {'name': 'cat', 'attributes': []},
            2: {'name': 'person', 'attributes': ['large']},
        }
    },
    'img_002': {
        'objects': {
            0: {'name': 'car', 'attributes': ['blue']},
            1: {'name': 'chair', 'attributes': ['red', 'large']},
        }
    },
}

# Create QA dict
qa = {
    'q_001': {
        'imageId': 'img_001',
        'annotations': {
            'answer': {0: 0, 1: 1},
            'fullAnswer': {0: 0},
            'question': {0: 2},
        }
    },
    'q_002': {
        'imageId': 'img_002',
        'annotations': {
            'answer': {0: 0},
            'fullAnswer': {0: 1},
            'question': {0: 0},
        }
    },
}

# Create relevant_objects_per_img_and_question
relevant_objects_per_img_and_question = {
    'img_001': {
        'q_001': {
            0: [0, 1, 2],
            1: [0, 1],
            2: [0, 2],
        }
    },
    'img_002': {
        'q_002': {
            0: [0, 1],
            1: [1],
        }
    },
}

ARGS = {
    'sg': sg,
    'qa': qa,
    'relevant_objects_per_img_and_question': relevant_objects_per_img_and_question,
    'data_dir': temp_dir,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
h5py, numpy — 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
get_FIscores_INF_semantic_matching Ran
ran; our contract draft was wrong, not the code
dreichCSL/TrueVG/preprocessing/truevg/prep_hints.py
code served (permissive licence) · get_code("ed7ea0c3706614fe")
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 os
import numpy as np
import json
import tempfile

# Create a temporary directory for test data
temp_dir = tempfile.mkdtemp()

# Create the required JSON file with GQA object info
gqa_data = {
    "img_1": {"objectsNum": 10},
    "img_2": {"objectsNum": 15},
    "img_3": {"objectsNum": 8},
}

json_path = os.path.join(temp_dir, 'output_gqa_detectron_objects_info.json')
with open(json_path, 'w') as f:
    json.dump(gqa_data, f)

# Build qa dict: maps question_id to question info with imageId
qa = {
    "q_1": {"imageId": "img_1", "question": "What is this?"},
    "q_2": {"imageId": "img_2", "question": "How many?"},
    "q_3": {"imageId": "img_3", "question": "Where?"},
}

# Build rel_obj dict: maps imageId -> qid -> list of relevant objects
rel_obj = {
    "img_1": {
        "q_1": [1, 2, 3],
    },
    "img_2": {
        "q_2": [5, 6],
    },
    "img_3": {
        "q_3": [0, 1],
    },
}

# Build infusion_database: maps imageId -> qid -> objid -> list of boxes
# Each box is a tuple (box_index, box_info)
infusion_database = {
    "img_1": {
        "q_1": {
            "obj_1": [(1, {"x": 10, "y": 20}), (2, {"x": 30, "y": 40})],
            "obj_2": [(3, {"x": 50, "y": 60})],
        }
    },
    "img_2": {
        "q_2": {
            "obj_5": [(5, {"x": 100, "y": 110})],
        }
    },
    "img_3": {
        "q_3": {
            "obj_0": [(0, {"x": 5, "y": 15})],
        }
    },
}

# Build arguments dict
ARGS = {
    "qa": qa,
    "rel_obj": rel_obj,
    "infusion_database": infusion_database,
    "data_dir": temp_dir,
    "max_objects": 100,
}
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 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
get_FIscores_INF_spatial_matching Ran
ran; our contract draft was wrong, not the code
dreichCSL/TrueVG/preprocessing/truevg/prep_hints.py
code served (permissive licence) · get_code("bbab318c93ed2fb0")
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 os
import json
import tempfile
import torch
import numpy as np

# Create a temporary directory for data
temp_dir = tempfile.mkdtemp()

# Create the required GQA detectron objects info JSON file
gqa_data = {
    "img_001": {"objectsNum": 50},
    "img_002": {"objectsNum": 30},
}
output_path = os.path.join(temp_dir, 'output_gqa_detectron_objects_info.json')
with open(output_path, 'w') as f:
    json.dump(gqa_data, f)

# Create qa dict with image IDs
qa = {
    "q_001": {"imageId": "img_001"},
    "q_002": {"imageId": "img_002"},
    "q_003": {"imageId": "img_001"},
}

# Create FIscores_spatial dict with numpy arrays indexed by question ID
torch.manual_seed(0)
FIscores_spatial = {
    "q_001": np.random.randn(100),
    "q_002": np.random.randn(100),
    "q_003": np.random.randn(100),
}

# Create infusion_database dict
# Structure: {img_id: {qid: {objid: [list of (box_idx, box_info) tuples]}}}
infusion_database = {
    "img_001": {
        "q_001": {
            "obj_1": [(5, "box_info_1"), (10, "box_info_2")],
            "obj_2": [(15, "box_info_3")],
        },
        "q_003": {
            "obj_3": [(100, "box_info_4")],
        },
    },
    "img_002": {
        "q_002": {
            "obj_4": [(8, "box_info_5")],
        },
    },
}

# Define ARGS dictionary
ARGS = {
    "qa": qa,
    "FIscores_spatial": FIscores_spatial,
    "infusion_database": infusion_database,
    "data_dir": temp_dir,
    "max_objects": 100,
}
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
get_iou Ran
ran, honoured the contract we drafted
dreichCSL/TrueVG/preprocessing/truevg/prep_evaluation.py
code served (permissive licence) · get_code("47aac06436736844")
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 float, 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.

# Setup code for get_iou function

# Define two bounding boxes as dictionaries
# bb1: bounding box 1 with coordinates (x1, y1) at top-left and (x2, y2) at bottom-right
bb1 = {
    'x1': 10,
    'y1': 20,
    'x2': 50,
    'y2': 80
}

# bb2: bounding box 2 with coordinates (x1, y1) at top-left and (x2, y2) at bottom-right
bb2 = {
    'x1': 30,
    'y1': 40,
    'x2': 70,
    'y2': 100
}

ARGS = {
    'bb1': bb1,
    'bb2': bb2
}
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 float, 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
get_overlap Ran
ran, honoured the contract we drafted
dreichCSL/TrueVG/preprocessing/truevg/prep_evaluation.py
code served (permissive licence) · get_code("77c2c6aa4beabb9b")
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 float, 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.

# Setup code for get_overlap function

# Define bounding boxes as dictionaries with x1, x2, y1, y2 keys
# bb1: object bounding box
bb1 = {
    'x1': 10,
    'y1': 20,
    'x2': 50,
    'y2': 60
}

# bb2: area bounding box that overlaps with bb1
bb2 = {
    'x1': 30,
    'y1': 40,
    'x2': 70,
    'y2': 80
}

ARGS = {
    'bb1': bb1,
    'bb2': bb2
}
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 float, 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
get_prediction_scores Ran
ran; our contract draft was wrong, not the code
dreichCSL/TrueVG/components/losses.py
code served (permissive licence) · get_code("1a250cde60dc9077")
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 int, shape (2, 1) 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)

# Construct logits: shape (batch_size=2, num_classes=4)
# After torch.max(logits, 1)[1], we get indices of shape (2,)
logits = torch.randn(2, 4)

# Construct labels: shape (batch_size=2, num_classes=4)
# The function does labels.gather(1, logits.unsqueeze(1))
# logits.unsqueeze(1) will have shape (2, 1) after max operation
# gather(1, indices) with indices shape (2, 1) requires labels to have compatible shape
labels = torch.randint(0, 10, (2, 4))

ARGS = {
    "logits": logits,
    "labels": labels,
}
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 int, shape (2, 1) — shape (2, 1) 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
get_truevg_qids Ran
ran; our contract draft was wrong, not the code
dreichCSL/TrueVG/preprocessing/truevg/prep_evaluation.py
code served (permissive licence) · get_code("f314ec8478f617c7")
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 torch

# Construct nested dictionary structures for the function parameters
# Based on the function logic:
# - rel_obj: dict[imgid] -> dict[qid] -> list (non-empty lists)
# - irrel_obj: dict[imgid] -> dict[qid] -> list (non-empty lists)
# - infusion_database: dict[imgid] -> dict[qid] -> dict[objid] -> list of boxes
#   where each box is [something, [class_id, ...], ...]

# Create rel_obj structure
rel_obj = {
    "img_1": {
        "q_1": [1, 2, 3],
        "q_2": [4, 5],
    },
    "img_2": {
        "q_3": [6],
    }
}

# Create irrel_obj structure
irrel_obj = {
    "img_1": {
        "q_1": [7, 8],
        "q_3": [9],
    },
    "img_2": {
        "q_2": [10],
        "q_4": [11],
    }
}

# Create infusion_database structure
# box format: [index, [class_id, ...], ...]
# Boxes with class_id != 11 or (class_id == 11 and index == 100) will be included in infused_qids
infusion_database = {
    "img_1": {
        "q_1": {
            "obj_1": [
                [50, [5, 6]],
                [100, [11, 12]],
            ],
        },
    },
    "img_2": {
        "q_3": {
            "obj_2": [
                [75, [11, 13]],
            ],
        },
    }
}

ARGS = {
    "rel_obj": rel_obj,
    "irrel_obj": irrel_obj,
    "infusion_database": infusion_database,
    "fpvg_mode": True,
}
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, 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
instance_bce_with_logits Ran
ran; our contract draft was wrong, not the code
dreichCSL/TrueVG/components/losses.py
code served (permissive licence) · get_code("2b00263b6afeb3ee")
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)

# Create a minimal config object with vqa_loss_type attribute
class OptConfig:
    def __init__(self, vqa_loss_type):
        self.vqa_loss_type = vqa_loss_type

# Test with sigmoid loss type
opt_sigmoid = OptConfig(vqa_loss_type='sigmoid')

# Create small tensors: batch_size=2, num_answers=4
batch_size = 2
num_answers = 4

# Logits: (batch_size, num_answers)
logits = torch.randn(batch_size, num_answers)

# Labels for sigmoid: (batch_size, num_answers) with values in [0, 1]
labels = torch.randint(0, 2, (batch_size, num_answers)).float()

ARGS = {
    'opt': opt_sigmoid,
    'logits': logits,
    'labels': labels
}
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

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 0 later) and 8 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
compute_score_with_logits 2 other papers
2 dated earlier
1912.02315 (2019-12)
12-in-1: Multi-Task Vision and Language Representation Learning
2109.11797 (2021-09)
CPT: Colorful Prompt Tuning for Pre-trained Vision-Language Models

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

Visual Grounding (VG) methods in Visual Question Answering (VQA) attempt to improve VQA performance by strengthening a model's reliance on question-relevant visual information. The presence of such relevant information in the visual input is typically assumed in training and testing. This assumption, however, is inherently flawed when dealing with imperfect image representations common in large-scale VQA, where the information carried by visual features frequently deviates from expected ground-truth contents. As a result, training and testing of VG-methods is performed with largely inaccurate data, which obstructs proper assessment of their potential benefits. In this study, we demonstrate that current evaluation schemes for VG-methods are problematic due to the flawed assumption of availability of relevant visual information. Our experiments show that these methods can be much more effective when evaluation conditions are corrected. Code is provided on GitHub 1 .

For agents

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

get_harvested_code_for_paper("2401.07803")
get_code_for_paper("2401.07803")
have("2401.07803")

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

curl https://syntology.ai/api/ran/2401.07803.json

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

[![Syntology run record](https://syntology.ai/api/ran/2401.07803.svg)](https://syntology.ai/paper/2401.07803)

Connect an agent — have() is free.