SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.11824 · ICML · 2026

REVIS: Sparse Latent Steering to Mitigate Object Hallucination in Large Vision-Language Models

Zhicong Huang, Jialin Wu, Han Shen, Wei Shi, Zhou Yang, Peigui Qi, Binghao Wang, Kunsheng Tang

arXiv · PDF · Open in the Atlas

Code that ran

11 of 17 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; 1 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 10 failures of our instrument; 6 no recorded run), measured 2026-11-22; 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/2602.11824.json · badge: /api/ran/2602.11824.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
antgroup/Revis canonical (classifier label)
classifier basis: census
11 ran · 6 no recorded run
FunctionStatusWhere it lives
calculate_metrics Ran
ran; our contract draft was wrong, not the code
antgroup/Revis/utils/mmvet_judge_only.py
code served (permissive licence) · get_code("64c97831bbe858d3")
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
from collections import Counter

# Create a temporary directory with the required mm-vet.json file
temp_dir = tempfile.mkdtemp()
mmvet_json_path = os.path.join(temp_dir, 'mm-vet.json')

# Create sample mm-vet.json data
mmvet_data = {
    "sample_1": {
        "capability": ["rec", "ocr"]
    },
    "sample_2": {
        "capability": ["know", "gen"]
    },
    "sample_3": {
        "capability": ["spat", "math"]
    },
    "sample_4": {
        "capability": ["rec"]
    }
}

with open(mmvet_json_path, 'w') as f:
    json.dump(mmvet_data, f)

# Create grade_results with sample scores
grade_results = {
    "sample_1": {"score": 95.5},
    "sample_2": {"score": 87.3},
    "sample_3": {"score": 92.1},
    "sample_4": {"score": 88.9},
    "sample_5": {"score": 90.0}  # sample_5 not in data, will be skipped
}

ARGS = {
    "mmvet_path": temp_dir,
    "grade_results": grade_results
}
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
get_llm_layers_module Ran
ran; our contract draft was wrong, not the code
antgroup/Revis/steering_utils/llm_layers.py
code served (permissive licence) · get_code("7da4de2c6f0bb6bc")
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 ModuleList. 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)

# Case 1: model.model.layers structure (standard LLM)
class SimpleTransformerLayer(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 4)
    
    def forward(self, x):
        return self.linear(x)

class StandardLLMModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.model = nn.Module()
        self.model.layers = nn.ModuleList([SimpleTransformerLayer() for _ in range(2)])
    
    def forward(self, x):
        return x

# Test with the standard structure
model = StandardLLMModel()

ARGS = {
    "model": model
}
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 ModuleList — a ModuleList output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ModuleList 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
get_pure_direction_with_pca Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
antgroup/Revis/steering_utils/vector_calculator.py
code served (permissive licence) · get_code("b5bdcadedce1d8bb")
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 1, dtype family float, shape (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
from sklearn.decomposition import PCA

torch.manual_seed(0)

# Create sample vectors: shape (batch_size, feature_dim)
# Using batch=2, dim=4 as specified
sample_vectors = torch.randn(2, 4)

ARGS = {
    "sample_vectors": sample_vectors,
    "n_components": 1,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
sklearn, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 1, dtype family float, shape (4,) — shape (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
grade_mmvet_results Ran
ran; our contract draft was wrong, not the code
antgroup/Revis/utils/mmvet_ori.py
code served (permissive licence) · get_code("5fe2830411127fe3")
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
from unittest.mock import patch, MagicMock

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

# Create mock mm-vet.json data
mmvet_data = {
    "sample_1": {
        "question": "What is x in the equation?",
        "answer": "-1 <AND> -5"
    },
    "sample_2": {
        "question": "Can you explain this meme?",
        "answer": "This meme is poking fun at the fact that the names of the countries Iceland and Greenland are misleading."
    }
}

# Create mock results data
results_data = {
    "sample_1": "x = -1 or x = -5",
    "sample_2": "The meme is about Iceland and Greenland names."
}

# Create directory structure
mmvet_path = os.path.join(temp_dir, "mmvet_data")
os.makedirs(mmvet_path, exist_ok=True)

# Write mock data files
mmvet_json_path = os.path.join(mmvet_path, "mm-vet.json")
with open(mmvet_json_path, 'w') as f:
    json.dump(mmvet_data, f)

result_file_path = os.path.join(temp_dir, "results.json")
with open(result_file_path, 'w') as f:
    json.dump(results_data, f)

output_dir_path = os.path.join(temp_dir, "output")

ARGS = {
    "mmvet_path": mmvet_path,
    "result_file": result_file_path,
    "output_dir": output_dir_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
requests, tqdm — 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
load_image_and_mask Ran
ran, honoured the contract we drafted
antgroup/Revis/steering_utils/data_loader.py
code served (permissive licence) · get_code("2e2e273f233576a7")
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 4, 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 os
import tempfile
from PIL import Image

# Create a temporary directory and a test image file
temp_dir = tempfile.mkdtemp()
temp_image_path = os.path.join(temp_dir, "test_image.png")

# Create a simple test image (RGB, small size)
test_image = Image.new('RGB', (100, 100), color=(255, 0, 0))
test_image.save(temp_image_path)

# Define ARGS dict with the image_path parameter
ARGS = {
    'image_path': temp_image_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 4, dtype family int — 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
load_steering_data Ran
ran; our contract draft was wrong, not the code
antgroup/Revis/steering_utils/data_loader.py
code served (permissive licence) · get_code("b5f4cdfb98eafc38")
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 json
import os
import tempfile

# Create a temporary JSON file with steering data
temp_dir = tempfile.mkdtemp()
json_path = os.path.join(temp_dir, 'steering_data.json')

# Create test data in list format
test_data = [
    {"id": 1, "label": "steer_left", "value": 0.5},
    {"id": 2, "label": "steer_right", "value": -0.3},
    {"id": 3, "label": "steer_straight", "value": 0.0},
]

# Write the test data to the temporary JSON file
with open(json_path, 'w', encoding='utf-8') as f:
    json.dump(test_data, f)

ARGS = {
    'json_path': 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 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
parse_pope_answer Ran
ran; our contract draft was wrong, not the code
antgroup/Revis/main_all_visual_only.py
code served (permissive licence) · get_code("22f3c8885d3d43b7")
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 string

def parse_pope_answer(text):
    text = text.lower().strip()
    text = text.translate(str.maketrans('', '', string.punctuation))
    words = text.split()
    has_yes = 'yes' in words
    has_no = 'no' in words
    if has_yes and has_no:
        return 'yes' if text.find('yes') < text.find('no') else 'no'
    if has_yes:
        return 'yes'
    if has_no:
        return 'no'
    return 'unknown'

ARGS = {
    'text': 'Yes, I think so!'
}
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 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
process_layer_calibration Ran
ran; our contract draft was wrong, not the code
antgroup/Revis/layer_select.py
code served (permissive licence) · get_code("2886b13f15990b8f")
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 tempfile

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

# Construct records with the required structure
# Each record needs 'risk' and 'category' keys
records = [
    {'risk': 0.1, 'category': 'TP'},
    {'risk': 0.15, 'category': 'TP'},
    {'risk': 0.2, 'category': 'TN'},
    {'risk': 0.25, 'category': 'TN'},
    {'risk': 0.6, 'category': 'FP'},
    {'risk': 0.65, 'category': 'FP'},
    {'risk': 0.7, 'category': 'FP'},
]

layer_idx = 0
output_dir = temp_dir
k = 75

ARGS = {
    'records': records,
    'layer_idx': layer_idx,
    'output_dir': output_dir,
    'k': k,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
matplotlib, numpy, seaborn — 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
qwen_api_call Ran
ran; our contract draft was wrong, not the code
antgroup/Revis/utils/mmvet_judge_only.py
code served (permissive licence) · get_code("b88e74963350fcf2")
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 json
import time

# Mock the requests module to avoid actual network calls
class MockResponse:
    def __init__(self):
        self.status_code = 200
    
    def raise_for_status(self):
        pass
    
    def json(self):
        return {
            'choices': [
                {
                    'message': {
                        'content': 'This is a mock response from the API.'
                    }
                }
            ]
        }

class MockRequests:
    @staticmethod
    def post(url, headers=None, data=None, timeout=None):
        return MockResponse()

import sys
from unittest.mock import MagicMock
sys.modules['requests'] = MockRequests()

# Now define the arguments for qwen_api_call
ARGS = {
    'system_prompt': 'You are a helpful AI assistant that provides concise and accurate answers.',
    'prompt': 'What is the capital of France?',
    'model': 'Qwen3-235B-A22B-Instruct-2507',
    'temperature': 0.25,
    'max_tokens': 512
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
requests — 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
read_jsonl Ran
ran; our contract draft was wrong, not the code
antgroup/Revis/utils/chair.py
code served (permissive licence) · get_code("098ce1dbb8a719bf")
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 json
import tempfile
import os

# Create a temporary JSONL file with valid JSON objects
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test.jsonl")

with open(temp_file, 'w', encoding='utf-8') as f:
    f.write('{"key": "value1", "number": 1}\n')
    f.write('{"key": "value2", "number": 2}\n')
    f.write('{"key": "value3", "number": 3}\n')

ARGS = {
    "file_path": temp_file
}
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
run_mmvet_inference Ran
ran; our contract draft was wrong, not the code
antgroup/Revis/utils/mmvet_ori.py
code served (permissive licence) · get_code("81483ccb3b201a30")
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
from pathlib import Path

# Create temporary directories and files for testing
temp_dir = tempfile.mkdtemp()
mmvet_path = os.path.join(temp_dir, 'mmvet_data')
os.makedirs(mmvet_path, exist_ok=True)

images_dir = os.path.join(mmvet_path, 'images')
os.makedirs(images_dir, exist_ok=True)

# Create a minimal mm-vet.json file
mmvet_data = {
    'sample_1': {
        'question': 'What is in this image?',
        'imagename': 'image1.jpg'
    },
    'sample_2': {
        'question': 'Describe the scene.',
        'imagename': 'image2.jpg'
    }
}

mmvet_json_path = os.path.join(mmvet_path, 'mm-vet.json')
with open(mmvet_json_path, 'w') as f:
    json.dump(mmvet_data, f)

# Create dummy image files
for img_name in ['image1.jpg', 'image2.jpg']:
    img_path = os.path.join(images_dir, img_name)
    # Create a minimal valid JPEG by writing binary data
    with open(img_path, 'wb') as f:
        # Minimal JPEG header
        f.write(b'\xff\xd8\xff\xe0\x00\x10JFIF\x00\x01\x01\x00\x00\x01\x00\x01\x00\x00')
        f.write(b'\xff\xd9')

# Create a simple inference function that takes image_path and question
def dummy_inference_fn(image_path, question):
    """Minimal inference function that returns a fixed response"""
    return f"Answer to '{question}' for image at {image_path}"

# Create output file path
result_file = os.path.join(temp_dir, 'results.json')

# Define ARGS dict with all required parameters
ARGS = {
    'inference_fn': dummy_inference_fn,
    'mmvet_path': mmvet_path,
    'result_file': result_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tqdm — 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
calculate_mme_metric Not yet run
unverified (harvested, no recorded run)
antgroup/Revis/utils/mme.py
code served (permissive licence) · get_code("d02c7b781becb0da")
How it ran

no invocation recorded.

eval_jsonl_file Not yet run
unverified (harvested, no recorded run)
antgroup/Revis/utils/mme.py
code served (permissive licence) · get_code("05907cd5f0aa2eeb")
How it ran

no invocation recorded.

get_model_inputs Not yet run
unverified (harvested, no recorded run)
antgroup/Revis/layer_select.py
code served (permissive licence) · get_code("475fb2362ae38138")
How it ran

no invocation recorded.

load_model_generic Not yet run
unverified (harvested, no recorded run)
antgroup/Revis/layer_select.py
code served (permissive licence) · get_code("dbb22e2435f1864b")
How it ran

no invocation recorded.

load_model_generic Not yet run
unverified (harvested, no recorded run)
antgroup/Revis/main_all_visual_only.py
code served (permissive licence) · get_code("8e970b80e8c8a067")
How it ran

no invocation recorded.

svd_flip Not yet run
unverified (harvested, no recorded run)
antgroup/Revis/steering_utils/pca.py
code served (permissive licence) · get_code("a5dbff5dd48042c6")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 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 17 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

Despite the advanced capabilities of Large Vision-Language Models (LVLMs), they frequently suffer from object hallucination. One reason is that visual features and pretrained textual representations often become intertwined in the deeper network layers. To address this, we propose RE-VIS, a training-free framework designed to explicitly re-activate this suppressed visual information. Rooted in latent space geometry, RE-VIS extracts the pure visual information vector via orthogonal projection and employs a calibrated strategy to perform sparse intervention only at the precise depth where suppression occurs. This surgical approach effectively restores visual information with minimal computational cost. Empirical evaluations on standard benchmarks demonstrate that REVIS reduces object hallucination rates by approximately 19% compared to state-of-the-art baselines, while preserving general reasoning capabilities. Code: https: //github.com/antgroup/Revis.

For agents

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

get_harvested_code_for_paper("2602.11824")
get_code_for_paper("2602.11824")
have("2602.11824")

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

curl https://syntology.ai/api/ran/2602.11824.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.11824.svg)](https://syntology.ai/paper/2602.11824)

Connect an agent — have() is free.