SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2510.04039 · ICML · 2025

GUI-Spotlight: Adaptive Iterative Focus Refinement for Enhanced GUI Visual Grounding

Caiwen Ding, Mingyi Hong, Nuo Xu, Ali Payani, Bin Lei, Yu Cao, Chunhua Liao

arXiv · PDF · Open in the Atlas

Code that ran

18 of 18 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; 16 failures of our instrument; 0 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/2510.04039.json · badge: /api/ran/2510.04039.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
bin123apple/GUI_Spotlight canonical (classifier label)
classifier basis: census
18 ran · 0 no recorded run
FunctionStatusWhere it lives
compute_iou Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
bin123apple/GUI_Spotlight/spotlight/reward/tool_rubric.py
code served (permissive licence) · get_code("4e08ab7b4f176f64")
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.

import torch

# Define ARGS dict with valid arguments for compute_iou
ARGS = {
    "pred_bbox": (50, 60, 100, 120),
    "gt_bbox": (55, 65, 105, 125),
    "img_size": (640, 480)
}
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 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
crop Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
bin123apple/GUI_Spotlight/spotlight/tools.py
code served (permissive licence) · get_code("3ce4db2cfef315c5")
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 os, io, cv2, numpy as np
from PIL import Image
from typing import Tuple, Optional, Union

# Create a simple test image
img_array = np.random.randint(0, 256, (100, 120, 3), dtype=np.uint8)
img = Image.fromarray(img_array, mode='RGB')

# Define valid crop coordinates
top_left = (10, 15)
bottom_right = (80, 90)

ARGS = {
    'img': img,
    'top_left': top_left,
    'bottom_right': bottom_right
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, cv2, numpy — 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
crop Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
bin123apple/GUI_Spotlight/inference.py
code served (permissive licence) · get_code("09aa6602ff0fe56f")
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 io
from PIL import Image
import numpy as np

# Create a small test image
img_array = np.random.randint(0, 256, (100, 100, 3), dtype=np.uint8)
img = Image.fromarray(img_array, mode='RGB')

# Define valid crop coordinates
top_left = (10, 10)
bottom_right = (60, 60)

ARGS = {
    'img': img,
    'top_left': top_left,
    'bottom_right': bottom_right
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, cv2, numpy — 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
crop Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
bin123apple/GUI_Spotlight/osworld_g_evaluation.py
code served (permissive licence) · get_code("753dc5bba606acbe")
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.

from PIL import Image
import io

# Create a simple test image
img = Image.new('RGB', (200, 150), color='red')

# Define valid crop arguments
ARGS = {
    'img': img,
    'top_left': (10, 10),
    'bottom_right': (100, 100)
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, cv2, numpy — 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
crop Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
bin123apple/GUI_Spotlight/screenspot_pro_evaluation.py
code served (permissive licence) · get_code("91214f85123a5c87")
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.

from PIL import Image, ImageDraw
import io

# Create a simple test image
img = Image.new('RGB', (200, 150), color='red')
draw = ImageDraw.Draw(img)
draw.rectangle([50, 40, 150, 120], fill='blue')

# Define valid crop coordinates
top_left = (50, 40)
bottom_right = (150, 120)

ARGS = {
    'img': img,
    'top_left': top_left,
    'bottom_right': bottom_right
}
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 — 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
encode_image Ran
ran; our contract draft was wrong, not the code
bin123apple/GUI_Spotlight/screenspot_pro_evaluation.py
code served (permissive licence) · get_code("78ee6b46cebd99ee")
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) 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 base64

# The function encode_image takes image_content (bytes) and base64 encodes it
# We need to provide bytes as the image_content parameter

image_content = b"This is sample image data in bytes format"

ARGS = {
    "image_content": image_content
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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
equals_reward_func Ran
ran, honoured the contract we drafted
bin123apple/GUI_Spotlight/spotlight/reward/rubric.py
code served (permissive licence) · get_code("264aee759c47ba3a")
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 torch.nn as nn

torch.manual_seed(0)

# Construct valid arguments for equals_reward_func
# The function expects:
# - completions: a list where each element c has c[0]['content'] accessible
# - answer: a list of expected answers
# - **kwargs: any additional keyword arguments (optional)

# Build completions as a list of lists of dicts with 'content' key
completions = [
    [{'content': 'correct_answer_1'}, {'content': 'extra_data'}],
    [{'content': 'wrong_answer'}, {'content': 'extra_data'}],
]

# Build answer as a list of expected strings
answer = ['correct_answer_1', 'correct_answer_1']

# Additional kwargs (empty dict is acceptable)
kwargs_dict = {}

ARGS = {
    'completions': completions,
    'answer': answer,
    **kwargs_dict
}
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
extract Ran
ran; our contract draft was wrong, not the code
bin123apple/GUI_Spotlight/spotlight/tools.py
code served (permissive licence) · get_code("af8d22af4cf5f3a9")
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 os
import io
import numpy as np
from PIL import Image

# Create a simple test image (100x100 RGB image)
test_image = Image.new('RGB', (100, 100), color=(255, 0, 0))

ARGS = {
    'img_input': test_image,
    'x_pos': 'center',
    'y_pos': 'center',
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, cv2, numpy — 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
extract Ran
ran; our contract draft was wrong, not the code
bin123apple/GUI_Spotlight/inference.py
code served (permissive licence) · get_code("67c6298ad2482e8f")
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 io
from PIL import Image
import numpy as np

# Create a small test image
width, height = 100, 100
test_image = Image.new('RGB', (width, height), color='red')

# Define arguments for the extract function
ARGS = {
    'img_input': test_image,
    'x_pos': 'center',
    'y_pos': 'center'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, cv2, numpy — 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
extract Ran
ran; our contract draft was wrong, not the code
bin123apple/GUI_Spotlight/osworld_g_evaluation.py
code served (permissive licence) · get_code("52342a2f125d6b3b")
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 io
import os
from PIL import Image
import numpy as np

# Create a simple test image
img_width = 100
img_height = 100
img_array = np.random.randint(0, 256, (img_height, img_width, 3), dtype=np.uint8)
test_image = Image.fromarray(img_array)

# Create a temporary file path for the test image
test_image_path = "/tmp/test_image.png"
test_image.save(test_image_path)

ARGS = {
    'img_input': test_image,
    'x_pos': 'center',
    'y_pos': 'center'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, cv2, numpy — 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
extract_first_image Ran
ran, honoured the contract we drafted
bin123apple/GUI_Spotlight/spotlight/reward/tool_rubric.py
code served (permissive licence) · get_code("d05b0d3e1cfb22cf")
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 PngImageFile. 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 base64
from io import BytesIO
from PIL import Image
import numpy as np

# Create a small test image
img = Image.new('RGB', (10, 10), color='red')
img_bytes = BytesIO()
img.save(img_bytes, format='PNG')
img_bytes.seek(0)
image_data = img_bytes.read()
b64_str = base64.b64encode(image_data).decode('utf-8')

# Construct a conversation list with an image
conv_list = [
    {
        'role': 'user',
        'content': [
            {
                'type': 'image_url',
                'image_url': {
                    'url': f'data:image/png;base64,{b64_str}'
                }
            }
        ]
    }
]

ARGS = {
    'conv_list': conv_list
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PngImageFile — a PngImageFile output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PngImageFile output the recorded ndim 3 and dtype family int 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
find_color Ran
ran; our contract draft was wrong, not the code
bin123apple/GUI_Spotlight/inference.py
code served (permissive licence) · get_code("a8300c75cc23637f")
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 io
import re
import base64
import copy
import os
import cv2
import numpy as np
from typing import Tuple, Optional, Union
from PIL import Image, ImageDraw

# Create a simple test image as PIL.Image.Image
test_image = Image.new('RGB', (300, 300), color=(100, 150, 200))
# Draw some colored patches to make it interesting
draw = ImageDraw.Draw(test_image)
draw.rectangle([50, 50, 100, 100], fill=(255, 0, 0))  # Red patch
draw.rectangle([150, 150, 200, 200], fill=(0, 255, 0))  # Green patch

# Define the arguments
ARGS = {
    'img_input': test_image,
    'target_rgb': (255, 0, 0)
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, cv2, numpy — 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
find_color Ran
ran; our contract draft was wrong, not the code
bin123apple/GUI_Spotlight/osworld_g_evaluation.py
code served (permissive licence) · get_code("fdf9d5aa2c45a4c2")
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 numpy as np
from PIL import Image
import cv2

# Create a small test image with PIL
test_image = Image.new('RGB', (100, 100), color=(73, 109, 137))
# Add a patch of the target color we'll search for
pixels = test_image.load()
for x in range(30, 50):
    for y in range(30, 50):
        pixels[x, y] = (255, 100, 50)

# Define the target RGB color to find
target_rgb = (255, 100, 50)

ARGS = {
    'img_input': test_image,
    'target_rgb': target_rgb,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, cv2, numpy — 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
format_tool_descriptions Ran
ran; our contract draft was wrong, not the code
bin123apple/GUI_Spotlight/spotlight/tools_envs/tool_env.py
code served (permissive licence) · get_code("894a351986ae8339")
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 torch.nn as nn

# Construct valid arguments for format_tool_descriptions

ARGS = {
    "schemas": [
        {
            "name": "calculate_sum",
            "description": "Adds two numbers together",
            "args": {
                "a": {
                    "description": "First number",
                    "type": "int"
                },
                "b": {
                    "description": "Second number",
                    "type": "int",
                    "default": 0
                }
            },
            "examples": [
                "calculate_sum(5, 3) returns 8",
                "calculate_sum(10) returns 10"
            ],
            "returns": "The sum of a and b"
        },
        {
            "name": "multiply_numbers",
            "description": "Multiplies two numbers",
            "args": {
                "x": {
                    "description": "First multiplicand",
                    "type": "float"
                },
                "y": {
                    "description": "Second multiplicand",
                    "type": "float",
                    "default": 1.0
                }
            },
            "examples": [
                "multiply_numbers(4, 5) returns 20",
                "multiply_numbers(3.5) returns 3.5"
            ],
            "returns": "The product of x and y"
        }
    ]
}
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
infer_schema_from_function Ran
ran; our contract draft was wrong, not the code
bin123apple/GUI_Spotlight/spotlight/tools_envs/tool_env.py
code served (permissive licence) · get_code("5948d381ce0e5cf4")
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 inspect
from typing import Callable, Dict, Any

def sample_function(x: int, y: str = "default") -> bool:
    """
    This is a sample function for testing schema inference.
    
    This function demonstrates how the schema inferrer works with
    a simple example containing parameters and a return type.
    
    Args:
        x: An integer parameter
        y: A string parameter with default value
    
    Returns:
        A boolean value indicating success
    
    Examples:
        sample_function(5)
        sample_function(10, "custom")
    """
    return True

ARGS = {
    "func": sample_function
}
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
iou Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
bin123apple/GUI_Spotlight/spotlight/tools.py
code served (permissive licence) · get_code("9dc039e3a4c8ba70")
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.

import torch
import numpy as np

torch.manual_seed(0)

# The function iou computes Intersection over Union for two bounding boxes
# Each bounding box is expected to be a sequence/array with 4 elements: [x1, y1, x2, y2]
# The function indexes into b1 and b2 using integer indices (0-3)
# and performs arithmetic operations on the values

# Create two bounding boxes as tuples or lists
# Format: [x1, y1, x2, y2] where x1,y1 is top-left and x2,y2 is bottom-right
b1 = (10, 20, 100, 150)  # box 1: x1=10, y1=20, x2=100, y2=150
b2 = (50, 60, 120, 180)  # box 2: x1=50, y1=60, x2=120, y2=180

ARGS = {
    "b1": b1,
    "b2": b2,
}
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 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
parse_crop_bbox_from_text Ran
ran; our contract draft was wrong, not the code
bin123apple/GUI_Spotlight/spotlight/reward/tool_rubric.py
code served (permissive licence) · get_code("5d9f44f31fe3f7ac")
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 re
import textwrap

# Test case 1: Valid crop bbox format
text_valid = """
<crop>
(
Image_0
,
(10, 20)
,
(110, 100)
)
</crop>
"""

# Test case 2: Valid crop with negative coordinates
text_with_negatives = """
<crop>
(
Photo_5
,
(-10, -20)
,
(50, 60)
)
</crop>
"""

# Test case 3: Invalid format - should return (None, None, None)
text_invalid = "This is just some random text without crop tags"

# Test case 4: Valid crop with different name and id
text_valid_2 = """
<crop>
(
Frame-1_42
,
(0, 0)
,
(640, 480)
)
</crop>
"""

ARGS = {
    "text": text_valid
}
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 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
sanitize_dialogs Ran
ran; our contract draft was wrong, not the code
bin123apple/GUI_Spotlight/screenspot_pro_evaluation.py
code served (permissive licence) · get_code("c91f63a1a917435c")
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. 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 copy
from typing import List, Dict, Any, Tuple

PLACEHOLDER = '<IMAGE>'

def sanitize_dialogs(dialogs: List[List[dict]], placeholder: str=PLACEHOLDER):
    safe = copy.deepcopy(dialogs)
    for dialog in safe:
        for msg in dialog:
            content = msg.get('content')
            if isinstance(content, list):
                for piece in content:
                    if piece.get('type') == 'image_url':
                        piece['image_url'] = placeholder
    return safe


# Setup code to construct valid arguments
dialogs_input = [
    [
        {
            'role': 'user',
            'content': [
                {
                    'type': 'text',
                    'text': 'What is in this image?'
                },
                {
                    'type': 'image_url',
                    'image_url': 'https://example.com/image1.jpg'
                }
            ]
        },
        {
            'role': 'assistant',
            'content': 'This is an image response.'
        }
    ],
    [
        {
            'role': 'user',
            'content': [
                {
                    'type': 'image_url',
                    'image_url': 'https://example.com/image2.jpg'
                },
                {
                    'type': 'text',
                    'text': 'Describe this.'
                }
            ]
        }
    ]
]

placeholder_input = '<IMAGE>'

ARGS = {
    'dialogs': dialogs_input,
    'placeholder': placeholder_input
}
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 — a list 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

Also present in other papers' code

Of 18 samples lifted from this paper's linked repositories, 18 carry a function identity and 0 do not. Of the 12 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, 1 later) and 11 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
encode_image 4 other papers
3 dated earlier, 1 later
2404.07972 (2024-04)
OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments
2407.10956 (2024-07)
Spider2-V: How Far Are Multimodal Agents From Automating Data Science and Engineering Workflows?
2505.21936 (2025-05)
RedTeamCUA: Realistic Adversarial Testing of Computer-Use Agents in Hybrid Web-OS Environments
and 1 more not named here

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

Multimodal large language models (MLLMs) have markedly expanded the competence of graphical user-interface (GUI) systems, propelling them beyond controlled simulations into complex, real-world environments across diverse platforms. However, practical usefulness is still bounded by the reliability of visual grounding, i.e., mapping textual references to exact on-screen elements. This limitation prevents the system from accurately performing pointer-level actions such as clicking or dragging. To address it, we introduce GUI-SPOTLIGHT-A model trained for image-grounded reasoning that dynamically invokes multiple specialized tools to iteratively narrow its focus to the relevant region of the screen, thereby substantially improving visual grounding accuracy. On the ScreenSpot-Pro benchmark, GUI-Spotlight trained with only 18.5K training samples achieves 52.8% accuracy, surpassing V2P-7B (50.6% with 9.6M training samples) and GTA-1-7B (50.1% with 1.56M training samples). Code is avaliable at https://github.com/bin123apple/GUI_Spotlight.

For agents

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

get_harvested_code_for_paper("2510.04039")
get_code_for_paper("2510.04039")
have("2510.04039")

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

curl https://syntology.ai/api/ran/2510.04039.json

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

[![Syntology run record](https://syntology.ai/api/ran/2510.04039.svg)](https://syntology.ai/paper/2510.04039)

Connect an agent — have() is free.