SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2401.11035 · AAAI · 2024

Image Safeguarding: Reasoning with Conditional Vision Language Model and Obfuscating Unsafe Content Counterfactually

Peyman Najafirad, Mazal Bethany, Brandon Wherry, Nishant Vishwamitra

arXiv · PDF · Open in the Atlas

Code that ran

12 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; 4 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 7 failures of our instrument; 4 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/2401.11035.json · badge: /api/ran/2401.11035.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
SecureAIAutonomyLab/ConditionalVLM canonical (classifier label)
classifier basis: census
12 ran · 4 no recorded run
FunctionStatusWhere it lives
cam_processor_for_segments Ran
ran, honoured the contract we drafted
SecureAIAutonomyLab/ConditionalVLM/CSE/src/cse/re_alogirthms.py
code served (permissive licence) · get_code("23cf764b6ef7887d")
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 2, 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 numpy as np

# Create small grayscale CAM output (4x3 matrix of floats)
grayscale_cam_output = np.array([
    [0.1, 0.2, 0.3],
    [0.4, 0.5, 0.6],
    [0.7, 0.8, 0.9],
    [0.2, 0.3, 0.4]
], dtype=np.float32)

# Create segments_slic with region labels (4x3 matrix of integers)
segments_slic = np.array([
    [0, 0, 1],
    [0, 1, 1],
    [2, 2, 2],
    [2, 3, 3]
], dtype=np.int32)

ARGS = {
    "grayscale_cam_output": grayscale_cam_output,
    "segments_slic": segments_slic
}
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 2, 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
conditional_prompt Ran
ran; our contract draft was wrong, not the code
SecureAIAutonomyLab/ConditionalVLM/data_scripts/parallel_evaluation_pass1_llama2.py
code served (permissive licence) · get_code("621d1dd76acb359f")
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 str. 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

# Define constants
NSFW = 'nsfw'
SELF_HARM = 'self_harm'
CYBERBULLYING = 'cyberbullying'

# Define ARGS dictionary with all required parameters
ARGS = {
    'llm_response': 'The image appears to show content that has been obscured or blurred, making it difficult to determine the exact nature of the content.',
    'datatype': NSFW,
    'question': 'Does this image contain any unsafe or obfuscated content?'
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
conditional_prompt Ran
ran; our contract draft was wrong, not the code
SecureAIAutonomyLab/ConditionalVLM/data_scripts/step1_generate_instruct_parallel.py
code served (permissive licence) · get_code("ccf528c84bf44e64")
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 str. 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

# Define the constants used in the function
NSFW = 'nsfw'
SELF_HARM = 'self_harm'
CYBERBULLYING = 'cyberbullying'

# Construct valid arguments for the conditional_prompt function
ARGS = {
    'question': 'What is happening in this image?',
    'datatype': NSFW,
    'alpha': 0.7,
    'is_safe': False
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
create_train_objs Ran
ran; our contract draft was wrong, not the code
SecureAIAutonomyLab/ConditionalVLM/CSE/src/cse/trainer.py
code served (permissive licence) · get_code("dc422009a87cede8")
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
import torch.nn.functional as F

torch.manual_seed(0)

ARGS = {
    "learning_rate": 0.001
}
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 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
csv_mask_to_numpy Ran
ran, honoured the contract we drafted
SecureAIAutonomyLab/ConditionalVLM/CSE/src/cse/bass_functions.py
code served (permissive licence) · get_code("5abba8fa7da75d50")
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 (12, 16) 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 tempfile
import os

# Create a temporary CSV file with integer data
temp_dir = tempfile.mkdtemp()
csv_path = os.path.join(temp_dir, "mask.csv")

# Create a small sample mask (3x4 matrix of integers)
sample_mask = np.array([
    [1, 0, 1, 0],
    [0, 1, 0, 1],
    [1, 1, 0, 0]
], dtype=int)

# Write to CSV file
np.savetxt(csv_path, sample_mask, delimiter=',', fmt='%d')

ARGS = {
    'csv_path': csv_path
}
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 (12, 16) — shape (12, 16) 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
csv_mask_to_numpy_SAM Ran
ran, honoured the contract we drafted
SecureAIAutonomyLab/ConditionalVLM/CSE/src/cse/bass_functions.py
code served (permissive licence) · get_code("a9bfb48da31a8195")
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 (3, 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 numpy as np
import tempfile
import os

# Create a temporary CSV file with mask data
temp_dir = tempfile.mkdtemp()
csv_path = os.path.join(temp_dir, "test_mask.csv")

# Create sample mask data as a 2D numpy array
mask_data = np.array([
    [0, 1, 0, 1],
    [1, 0, 1, 0],
    [0, 1, 1, 0]
], dtype=int)

# Write to CSV file
np.savetxt(csv_path, mask_data, delimiter=',', fmt='%d')

ARGS = {
    "csv_path": csv_path
}
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 (3, 4) — shape (3, 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
get_feature_masks Ran
ran, honoured the contract we drafted
SecureAIAutonomyLab/ConditionalVLM/CSE/src/cse/re_alogirthms_unsafe.py
code served (permissive licence) · get_code("754eb64511d7672f")
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 3, dtype family int; 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 numpy as np

# Analyze the function:
# - image: passed but not used in the function body
# - attributions: iterated over, each element i is compared with segments_slic using np.where
# - segments_slic: used in np.where comparison with elements from attributions

# Create small numpy arrays with compatible shapes
# attributions should be iterable of arrays that can be compared with segments_slic
# Let's make them all the same shape (3, 4) for a small 2D grid

segments_slic = np.array([
    [0, 1, 1, 2],
    [0, 1, 2, 2],
    [3, 3, 2, 2]
], dtype=np.int32)

# attributions is a list of arrays with same shape as segments_slic
attributions = [
    np.array([
        [0, 1, 1, 2],
        [0, 1, 2, 2],
        [3, 3, 2, 2]
    ], dtype=np.int32),
    np.array([
        [1, 0, 1, 2],
        [0, 1, 2, 3],
        [3, 3, 2, 1]
    ], dtype=np.int32)
]

# image is not used but we need to provide it
image = np.zeros((3, 4, 3), dtype=np.uint8)

ARGS = {
    "image": image,
    "attributions": attributions,
    "segments_slic": segments_slic
}
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 3, dtype family int — 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_first_word Ran
ran; our contract draft was wrong, not the code
SecureAIAutonomyLab/ConditionalVLM/data_scripts/step2_human_verification.py
code served (permissive licence) · get_code("2288a760079fcb43")
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 str. 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

def get_first_word(s):
    words = s.split()
    first_word = words[0]
    first_word = first_word.strip(',')
    first_word = first_word.lower()
    return first_word

ARGS = {
    "s": "Hello, world!"
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
load_json_file Ran
ran; our contract draft was wrong, not the code
SecureAIAutonomyLab/ConditionalVLM/data_scripts/parallel_evaluation_pass1_llama2.py
code served (permissive licence) · get_code("df6d89d388fe4133")
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 json
from pathlib import Path
import tempfile
import os

# Create a temporary JSON file with valid test data
temp_dir = tempfile.mkdtemp()
test_json_path = Path(temp_dir) / "test_data.json"

test_data = {
    "key1": "value1",
    "key2": [1, 2, 3],
    "key3": {"nested": "object"}
}

with open(test_json_path, 'w') as f:
    json.dump(test_data, f)

ARGS = {
    "path_to_json": test_json_path
}
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
parse_response Ran
ran, violated the contract we drafted
SecureAIAutonomyLab/ConditionalVLM/data_scripts/step2_human_verification.py
code served (permissive licence) · get_code("d7840e6c2b484338")
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 bool, 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 numpy as np
from typing import List, Dict

# Define constants
LLM_ANSWERS = 'llm_answers'
ANSWERS = 'answers'

# Construct the data structure
# data is a List[Dict] where each dict contains 'llm_answers' and 'answers' keys
# Each of those keys maps to a list of responses
data = [
    {
        LLM_ANSWERS: ['llm_response_0', 'llm_response_1', 'llm_response_2'],
        ANSWERS: ['answer_0', 'answer_1', 'answer_2']
    },
    {
        LLM_ANSWERS: ['llm_response_0_alt', 'llm_response_1_alt', 'llm_response_2_alt'],
        ANSWERS: ['answer_0_alt', 'answer_1_alt', 'answer_2_alt']
    }
]

# num: must be in range 1-4
num = 1

# i: index into the responses list (must be valid for data[j])
i = 0

# j: index into the data list
j = 0

ARGS = {
    'num': num,
    'data': data,
    'i': i,
    'j': j
}
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 bool, 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
segmentation_info Ran
ran; our contract draft was wrong, not the code
SecureAIAutonomyLab/ConditionalVLM/CSE/src/cse/re_alogirthms.py
code served (permissive licence) · get_code("df4461bb26ce6c37")
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)

# Create a small image tensor
# The function calls: image.detach().cpu().squeeze().numpy()
# So we need a torch tensor with some spatial dimensions
image = torch.randn(1, 3, 8, 8)  # batch=1, channels=3, height=8, width=8

num_segments = 4
compactness = 10.0

ARGS = {
    "image": image,
    "num_segments": num_segments,
    "compactness": compactness,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, skimage, 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
splice_data_based_on_rank Ran
ran; our contract draft was wrong, not the code
SecureAIAutonomyLab/ConditionalVLM/data_scripts/parallel_evaluation_pass1_llama2.py
code served (permissive licence) · get_code("faf3a939d9923799")
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
import numpy as np
from typing import List, Dict

torch.manual_seed(0)

# Create sample data - a list of dictionaries
data = [
    {"id": 0, "value": "item_0"},
    {"id": 1, "value": "item_1"},
    {"id": 2, "value": "item_2"},
    {"id": 3, "value": "item_3"},
    {"id": 4, "value": "item_4"},
    {"id": 5, "value": "item_5"},
]

# Set up arguments for the function
ARGS = {
    "rank": 0,
    "data": data,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
accelerate, numpy, 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
create_dataloaders_MNIST Not yet run
unverified (harvested, no recorded run)
SecureAIAutonomyLab/ConditionalVLM/CSE/src/cse/trainer.py
code served (permissive licence) · get_code("232d178fceae3ade")
How it ran

no invocation recorded.

create_dataloaders_MNIST_fashion Not yet run
unverified (harvested, no recorded run)
SecureAIAutonomyLab/ConditionalVLM/CSE/src/cse/trainer.py
code served (permissive licence) · get_code("65ba7c82b3357718")
How it ran

no invocation recorded.

main Not yet run
unverified (harvested, no recorded run)
SecureAIAutonomyLab/ConditionalVLM/CSE/src/cse/__main__.py
code served (permissive licence) · get_code("2684e3caff3c9539")
How it ran

no invocation recorded.

safe_or_unsafe_and_dataset_type Not yet run
unverified (harvested, no recorded run)
SecureAIAutonomyLab/ConditionalVLM/data_scripts/step0_instruct_data_gen.py
code served (permissive licence) · get_code("a0ef73012388f178")
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, 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 16 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

Social media platforms are being increasingly used by malicious actors to share unsafe content, such as images depicting sexual activity, cyberbullying, and self-harm. Consequently, major platforms use artificial intelligence (AI) and human moderation to obfuscate such images to make them safer. Two critical needs for obfuscating unsafe images is that an accurate rationale for obfuscating image regions must be provided, and the sensitive regions should be obfuscated (e.g. blurring) for users' safety. This process involves addressing two key problems: (1) the reason for obfuscating unsafe images demands the platform to provide an accurate rationale that must be grounded in unsafe imagespecific attributes, and (2) the unsafe regions in the image must be minimally obfuscated while still depicting the safe regions. In this work, we address these key issues by first performing visual reasoning by designing a visual reasoning model (VLM) conditioned on pre-trained unsafe image classifiers to provide an accurate rationale grounded in unsafe image attributes, and then proposing a counterfactual explanation algorithm that minimally identifies and obfuscates unsafe regions for safe viewing, by first utilizing an unsafe image classifier attribution matrix to guide segmentation for a more optimal subregion segmentation followed by an informed greedy search to determine the minimum number of subregions required to modify the classifier's output based on attribution score. Extensive experiments on uncurated data from social networks emphasize the efficacy of our proposed method. We make our code available at: https://github.com/SecureAIAutonomyLab/ConditionalVLM

For agents

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

get_harvested_code_for_paper("2401.11035")
get_code_for_paper("2401.11035")
have("2401.11035")

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

curl https://syntology.ai/api/ran/2401.11035.json

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

[![Syntology run record](https://syntology.ai/api/ran/2401.11035.svg)](https://syntology.ai/paper/2401.11035)

Connect an agent — have() is free.