SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2202.06934 · 2022

Slicing Aided Hyper Inference and Fine-tuning for Small Object Detection

arXiv · PDF · Open in the Atlas

Code that ran

18 of 19 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 17 have no contract verdict; 1 failure of our instrument; 1 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/2202.06934.json · badge: /api/ran/2202.06934.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
obss/sahi canonical (classifier label)
classifier basis: classifier_2025
1 ran · 1 no recorded run
fcakyon/sahi-benchmark canonical (classifier label)
classifier basis not recorded on the link
1 ran · 0 no recorded run
Resham-Sundar/sahi-yolox community (classifier label)
classifier basis not recorded on the link
16 ran · 0 no recorded run
FunctionStatusWhere it lives
annotation_inside_slice Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
obss/sahi/sahi/slicing.py
code served (permissive licence) · get_code("52989f234436027c")
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.

# SETUP CODE

# Annotation in COCO format with bbox [x, y, width, height]
annotation = {
    'bbox': [10, 20, 50, 30]  # x=10, y=20, width=50, height=30
}

# Slice bbox in format [x_min, y_min, x_max, y_max]
slice_bbox = [5, 15, 100, 100]

ARGS = {
    'annotation': annotation,
    'slice_bbox': slice_bbox
}
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 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
annotation_inside_slice Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/slicing.py
code served (permissive licence) · get_code("e257047ab1fda46c")
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

# Construct annotation in COCO format
# The function accesses annotation['bbox'] which should be [left, top, width, height]
annotation = {
    'bbox': [10, 20, 30, 40]  # left=10, top=20, width=30, height=40
}

# Construct slice_bbox in format [x_min, y_min, x_max, y_max]
# This should create overlap with the annotation for testing
slice_bbox = [15, 25, 50, 70]  # x_min=15, y_min=25, x_max=50, y_max=70

ARGS = {
    'annotation': annotation,
    'slice_bbox': slice_bbox
}
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
batched_greedy_nmm Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/postprocess/combine.py
code served (permissive licence) · get_code("b54053e66ee738d8")
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 torch

torch.manual_seed(0)

# Create a small tensor with bounding boxes, scores, and category IDs
# Shape: [num_boxes, 6] where columns are [x1, y1, x2, y2, score, category_id]
object_predictions_as_tensor = torch.tensor([
    [10.0, 10.0, 20.0, 20.0, 0.9, 1.0],
    [11.0, 11.0, 21.0, 21.0, 0.8, 1.0],
    [50.0, 50.0, 60.0, 60.0, 0.85, 2.0],
    [51.0, 51.0, 61.0, 61.0, 0.7, 2.0],
], dtype=torch.float32)

ARGS = {
    'object_predictions_as_tensor': object_predictions_as_tensor,
    'match_metric': 'IOU',
    'match_threshold': 0.5,
}
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 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
batched_nms Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/postprocess/combine.py
code served (permissive licence) · get_code("8ed0517756dd1b38")
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 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 torch

torch.manual_seed(0)

# Create a small predictions tensor with shape [num_boxes, 6]
# Columns: [x1, y1, x2, y2, score, category_id]
predictions = torch.tensor([
    [10.0, 10.0, 20.0, 20.0, 0.9, 0.0],
    [12.0, 12.0, 22.0, 22.0, 0.8, 0.0],
    [50.0, 50.0, 60.0, 60.0, 0.85, 1.0],
    [51.0, 51.0, 61.0, 61.0, 0.75, 1.0],
], dtype=torch.float32)

ARGS = {
    'predictions': predictions,
    'match_metric': 'IOU',
    'match_threshold': 0.5,
}
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 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
calculate_box_union Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/postprocess/utils.py
code served (permissive licence) · get_code("686cde694bfeddb7")
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 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
from typing import List, Union

def calculate_box_union(box1: Union[List[int], np.ndarray], box2: Union[List[int], np.ndarray]) -> List[int]:
    """
    Args:
        box1 (List[int]): [x1, y1, x2, y2]
        box2 (List[int]): [x1, y1, x2, y2]
    """
    box1 = np.array(box1)
    box2 = np.array(box2)
    left_top = np.minimum(box1[:2], box2[:2])
    right_bottom = np.maximum(box1[2:], box2[2:])
    return list(np.concatenate((left_top, right_bottom)))

ARGS = {
    "box1": [10, 20, 50, 80],
    "box2": [30, 15, 70, 100]
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, 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
fix_full_shape_list Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/utils/compatibility.py
code served (permissive licence) · get_code("4b5d7326bb4f1646")
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 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 torch
import numpy as np

torch.manual_seed(0)

# The function takes one parameter: full_shape_list
# It checks if full_shape_list is not None and if the first element is an int or float
# If so, it wraps the entire list in another list
# Otherwise it returns full_shape_list as-is

# Test case 1: A list of integers/floats (should be wrapped)
full_shape_list_1 = [2, 3, 4, 5]

ARGS = {
    "full_shape_list": full_shape_list_1
}
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 list, ndim 2, 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
fix_shift_amount_list Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/utils/compatibility.py
code served (permissive licence) · get_code("abc6663a9ae9fce0")
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 torch
import numpy as np

torch.manual_seed(0)

# The function fix_shift_amount_list takes a shift_amount_list parameter
# It checks if the first element is an int or float
# If so, it wraps the entire list in another list
# Otherwise it returns the list as-is

# Test case 1: List of numbers (should be wrapped)
shift_amount_list_case1 = [1.5, 2.0, 3.5]

# This will be wrapped into [[1.5, 2.0, 3.5]]

ARGS = {
    "shift_amount_list": shift_amount_list_case1
}
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 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
get_ordered_image_name_list Ran
ran on a synthesized input
fcakyon/sahi-benchmark/xview/xview_to_coco.py
code served (permissive licence) · get_code("d0520a0e6473e8cc")
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
from typing import Dict, List

def get_ordered_image_name_list(image_name_to_annotation_ind: Dict):
    image_name_list: List[str] = list(image_name_to_annotation_ind.keys())

    def get_image_ind(image_name: str):
        return int(image_name.split('.')[0])
    image_name_list.sort(key=get_image_ind)
    return image_name_list


# Create a dictionary where keys are image names (strings with numeric prefixes)
# and values are annotation indices
image_name_to_annotation_ind = {
    "001.jpg": 0,
    "005.png": 1,
    "003.bmp": 2,
    "002.gif": 3,
}

ARGS = {
    "image_name_to_annotation_ind": image_name_to_annotation_ind,
}
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
get_slice_bboxes Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/slicing.py
code served (permissive licence) · get_code("0e3c82c0dee22fd7")
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 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 torch
import numpy as np

# Define ARGS with valid arguments for get_slice_bboxes
ARGS = {
    "image_height": 1024,
    "image_width": 1536,
    "slice_height": 512,
    "slice_width": 512,
    "overlap_height_ratio": 0.2,
    "overlap_width_ratio": 0.2,
}
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 list, ndim 2, 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
merge Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/utils/coco.py
code served (permissive licence) · get_code("4d47f766c74c1466")
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 copy
import numpy as np

def create_minimal_coco_dict(num_images=2, num_annotations=3, categories=None):
    """Helper to create a minimal valid COCO dict"""
    if categories is None:
        categories = [
            {'id': 1, 'name': 'car', 'supercategory': 'car'},
            {'id': 2, 'name': 'human', 'supercategory': 'human'},
        ]
    
    images = [
        {'id': i, 'file_name': f'image_{i}.jpg', 'height': 480, 'width': 640}
        for i in range(1, num_images + 1)
    ]
    
    annotations = [
        {
            'id': i,
            'image_id': (i % num_images) + 1,
            'category_id': (i % len(categories)) + 1,
            'area': 100.0,
            'bbox': [10, 10, 50, 50],
            'iscrowd': 0
        }
        for i in range(1, num_annotations + 1)
    ]
    
    return {
        'images': images,
        'annotations': annotations,
        'categories': categories
    }

# Create two distinct COCO dicts
coco_dict1 = create_minimal_coco_dict(
    num_images=2,
    num_annotations=3,
    categories=[
        {'id': 1, 'name': 'car', 'supercategory': 'car'},
        {'id': 2, 'name': 'human', 'supercategory': 'human'},
    ]
)

coco_dict2 = create_minimal_coco_dict(
    num_images=2,
    num_annotations=3,
    categories=[
        {'id': 1, 'name': 'car', 'supercategory': 'car'},
        {'id': 2, 'name': 'human', 'supercategory': 'human'},
    ]
)

# Optional desired_name2id for filtering/reordering
desired_name2id = {
    'car': 1,
    'human': 2
}

ARGS = {
    'coco_dict1': coco_dict1,
    'coco_dict2': coco_dict2,
    'desired_name2id': desired_name2id
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
merge_from_list Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/utils/coco.py
code served (permissive licence) · get_code("4ab401060d4dd491")
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 copy
import numpy as np

# Create sample COCO dictionary 1
coco_dict1 = {
    'images': [
        {'id': 1, 'file_name': 'img1.jpg', 'height': 480, 'width': 640},
        {'id': 2, 'file_name': 'img2.jpg', 'height': 480, 'width': 640},
    ],
    'annotations': [
        {'id': 1, 'image_id': 1, 'category_id': 1, 'area': 100, 'bbox': [10, 20, 50, 60], 'iscrowd': 0},
        {'id': 2, 'image_id': 1, 'category_id': 2, 'area': 200, 'bbox': [100, 100, 100, 100], 'iscrowd': 0},
        {'id': 3, 'image_id': 2, 'category_id': 1, 'area': 150, 'bbox': [30, 40, 60, 70], 'iscrowd': 0},
    ],
    'categories': [
        {'id': 1, 'name': 'human', 'supercategory': 'person'},
        {'id': 2, 'name': 'car', 'supercategory': 'vehicle'},
    ]
}

# Create sample COCO dictionary 2
coco_dict2 = {
    'images': [
        {'id': 1, 'file_name': 'img3.jpg', 'height': 480, 'width': 640},
    ],
    'annotations': [
        {'id': 1, 'image_id': 1, 'category_id': 1, 'area': 120, 'bbox': [50, 50, 80, 80], 'iscrowd': 0},
        {'id': 2, 'image_id': 1, 'category_id': 3, 'area': 300, 'bbox': [200, 200, 150, 150], 'iscrowd': 0},
    ],
    'categories': [
        {'id': 1, 'name': 'human', 'supercategory': 'person'},
        {'id': 3, 'name': 'big_vehicle', 'supercategory': 'vehicle'},
    ]
}

# Create a list of COCO dicts
coco_dict_list = [coco_dict1, coco_dict2]

# Define desired category mapping
desired_name2id = {
    'human': 1,
    'car': 2,
    'big_vehicle': 3,
}

ARGS = {
    'coco_dict_list': coco_dict_list,
    'desired_name2id': desired_name2id,
    'verbose': 1,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
nms Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/postprocess/combine.py
code served (permissive licence) · get_code("272238bc5ffeef61")
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 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 torch

torch.manual_seed(0)

# Create a small tensor of predictions with shape [num_boxes, 5]
# Each row: [x1, y1, x2, y2, score]
predictions = torch.tensor([
    [10.0, 10.0, 50.0, 50.0, 0.9],
    [15.0, 15.0, 55.0, 55.0, 0.8],
    [100.0, 100.0, 150.0, 150.0, 0.85]
], dtype=torch.float32)

ARGS = {
    'predictions': predictions,
    'match_metric': 'IOU',
    'match_threshold': 0.5
}
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 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
object_prediction_list_to_numpy Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/postprocess/utils.py
code served (permissive licence) · get_code("1c10c8ef13619529")
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 float, shape (2, 6) 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 collections.abc import Sequence

torch.manual_seed(0)

# Define minimal stand-in classes for the object prediction structure
class BBox:
    def __init__(self, coords):
        self.coords = coords
    
    def to_voc_bbox(self):
        return self.coords

class Score:
    def __init__(self, value):
        self.value = value

class Category:
    def __init__(self, id):
        self.id = id

class ObjectPrediction:
    def __init__(self, bbox, score, category):
        self.bbox = bbox
        self.score = score
        self.category = category
    
    def tolist(self):
        return self

# Build a minimal ObjectPredictionList with 2 predictions
bbox1 = BBox([10, 20, 30, 40])
score1 = Score(0.95)
category1 = Category(1)
pred1 = ObjectPrediction(bbox1, score1, category1)

bbox2 = BBox([50, 60, 100, 120])
score2 = Score(0.75)
category2 = Category(2)
pred2 = ObjectPrediction(bbox2, score2, category2)

class ObjectPredictionList(Sequence):
    def __init__(self, list):
        self.list = list
        super().__init__()

    def __getitem__(self, i):
        if torch.is_tensor(i) or isinstance(i, np.ndarray):
            i = i.tolist()
        if isinstance(i, int):
            return ObjectPredictionList([self.list[i]])
        elif isinstance(i, (tuple, list)):
            accessed_mapping = map(self.list.__getitem__, i)
            return ObjectPredictionList(list(accessed_mapping))
        else:
            raise NotImplementedError(f'{type(i)}')

    def __setitem__(self, i, elem):
        if torch.is_tensor(i) or isinstance(i, np.ndarray):
            i = i.tolist()
        if isinstance(i, int):
            self.list[i] = elem
        elif isinstance(i, (tuple, list)):
            if len(i) != len(elem):
                raise ValueError()
            if isinstance(elem, ObjectPredictionList):
                for ind, el in enumerate(elem.list):
                    self.list[i[ind]] = el
            else:
                for ind, el in enumerate(elem):
                    self.list[i[ind]] = el
        else:
            raise NotImplementedError(f'{type(i)}')

    def __len__(self):
        return len(self.list)

    def __str__(self):
        return str(self.list)

    def extend(self, object_prediction_list):
        self.list.extend(object_prediction_list.list)

    def tolist(self):
        if len(self.list) == 1:
            return self.list[0]
        else:
            return self.list

object_prediction_list = ObjectPredictionList([pred1, pred2])

ARGS = {
    'object_prediction_list': object_prediction_list
}
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 ndarray, ndim 2, dtype family float, shape (2, 6) — shape (2, 6) 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
object_prediction_list_to_torch Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/postprocess/utils.py
code served (permissive licence) · get_code("2de419bb83449d6e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (2, 6) 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 collections.abc import Sequence

torch.manual_seed(0)

# Stand-in classes for BBox, Score, and Category
class BBox:
    def __init__(self, x1, y1, x2, y2):
        self.x1 = x1
        self.y1 = y1
        self.x2 = x2
        self.y2 = y2
    
    def to_voc_bbox(self):
        return [self.x1, self.y1, self.x2, self.y2]

class Score:
    def __init__(self, value):
        self.value = value

class Category:
    def __init__(self, id):
        self.id = id

class ObjectPrediction:
    def __init__(self, bbox, score, category):
        self.bbox = bbox
        self.score = score
        self.category = category
    
    def tolist(self):
        return self

# Minimal ObjectPredictionList implementation
class ObjectPredictionList(Sequence):
    def __init__(self, list):
        self.list = list
        super().__init__()

    def __getitem__(self, i):
        if torch.is_tensor(i) or isinstance(i, np.ndarray):
            i = i.tolist()
        if isinstance(i, int):
            return ObjectPredictionList([self.list[i]])
        elif isinstance(i, (tuple, list)):
            accessed_mapping = map(self.list.__getitem__, i)
            return ObjectPredictionList(list(accessed_mapping))
        else:
            raise NotImplementedError(f'{type(i)}')

    def __setitem__(self, i, elem):
        if torch.is_tensor(i) or isinstance(i, np.ndarray):
            i = i.tolist()
        if isinstance(i, int):
            self.list[i] = elem
        elif isinstance(i, (tuple, list)):
            if len(i) != len(elem):
                raise ValueError()
            if isinstance(elem, ObjectPredictionList):
                for ind, el in enumerate(elem.list):
                    self.list[i[ind]] = el
            else:
                for ind, el in enumerate(elem):
                    self.list[i[ind]] = el
        else:
            raise NotImplementedError(f'{type(i)}')

    def __len__(self):
        return len(self.list)

    def __str__(self):
        return str(self.list)

    def extend(self, object_prediction_list):
        self.list.extend(object_prediction_list.list)

    def tolist(self):
        if len(self.list) == 1:
            return self.list[0]
        else:
            return self.list

def object_prediction_list_to_torch(object_prediction_list: ObjectPredictionList) -> torch.tensor:
    """
    Returns:
        torch.tensor of size N x [x1, y1, x2, y2, score, category_id]
    """
    num_predictions = len(object_prediction_list)
    torch_predictions = torch.zeros([num_predictions, 6], dtype=torch.float32)
    for ind, object_prediction in enumerate(object_prediction_list):
        torch_predictions[ind, :4] = torch.tensor(object_prediction.tolist().bbox.to_voc_bbox(), dtype=torch.int32)
        torch_predictions[ind, 4] = object_prediction.tolist().score.value
        torch_predictions[ind, 5] = object_prediction.tolist().category.id
    return torch_predictions

# Create sample predictions
pred1 = ObjectPrediction(
    bbox=BBox(10, 20, 30, 40),
    score=Score(0.95),
    category=Category(1)
)
pred2 = ObjectPrediction(
    bbox=BBox(50, 60, 70, 80),
    score=Score(0.87),
    category=Category(2)
)

object_prediction_list = ObjectPredictionList([pred1, pred2])

ARGS = {
    "object_prediction_list": object_prediction_list
}
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 Tensor, ndim 2, dtype family float, shape (2, 6) — shape (2, 6) 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
read_image Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/utils/cv.py
code served (permissive licence) · get_code("7a2ae5c4bf39dd82")
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 3, dtype family int, shape (100, 100, 3) 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 cv2
import numpy as np
import tempfile
import os

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

# Create a simple test image (100x100 BGR image)
test_image = np.zeros((100, 100, 3), dtype=np.uint8)
test_image[:, :, 0] = 50   # Blue channel
test_image[:, :, 1] = 100  # Green channel
test_image[:, :, 2] = 150  # Red channel

cv2.imwrite(image_path, test_image)

ARGS = {
    "image_path": image_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family int, shape (100, 100, 3) — shape (100, 100, 3) 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
read_image_as_pil Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/utils/cv.py
code served (permissive licence) · get_code("c8c3e5a8345b053a")
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 Image. 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

# Create a test image as a PIL Image
test_image_pil = Image.new('RGB', (100, 100), color='red')

# Create a test numpy array (height, width, channels)
test_image_array = np.random.randint(0, 256, size=(100, 100, 3), dtype=np.uint8)

# Define arguments for read_image_as_pil
ARGS = {
    "image": test_image_pil,
    "exif_fix": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy, requests, skimage — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Image — a Image output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Image 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
read_large_image Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/utils/cv.py
code served (permissive licence) · get_code("91ba8065c40815aa")
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 cv2
import numpy as np
import tempfile
import os

# Create a temporary image file for testing
def create_test_image():
    # Create a simple test image (10x10 RGB image)
    img_array = np.random.randint(0, 256, (10, 10, 3), dtype=np.uint8)
    
    # Write to a temporary file
    with tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp:
        tmp_path = tmp.name
    
    # Use cv2 to write the image in BGR format
    cv2.imwrite(tmp_path, cv2.cvtColor(img_array, cv2.COLOR_RGB2BGR))
    
    return tmp_path

# Create the test image
test_image_path = create_test_image()

ARGS = {
    'image_path': test_image_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy, skimage — 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
update_categories Ran
ran on a synthesized input
Resham-Sundar/sahi-yolox/sahi/utils/coco.py
code served (permissive licence) · get_code("ece1ba05d166226f")
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 copy

# Construct valid arguments for update_categories function

# desired_name2id: dict mapping category names to desired IDs
desired_name2id = {
    "big_vehicle": 1,
    "car": 2,
    "human": 3
}

# coco_dict: COCO formatted dictionary with images, annotations, and categories
coco_dict = {
    'images': [
        {'id': 1, 'file_name': 'image1.jpg', 'height': 480, 'width': 640},
        {'id': 2, 'file_name': 'image2.jpg', 'height': 480, 'width': 640},
    ],
    'annotations': [
        {
            'id': 1,
            'image_id': 1,
            'category_id': 10,  # Maps to "big_vehicle" (desired id 1)
            'bbox': [10, 10, 100, 100],
            'area': 10000,
            'iscrowd': 0
        },
        {
            'id': 2,
            'image_id': 1,
            'category_id': 11,  # Maps to "car" (desired id 2)
            'bbox': [150, 150, 80, 80],
            'area': 6400,
            'iscrowd': 0
        },
        {
            'id': 3,
            'image_id': 2,
            'category_id': 12,  # Maps to "human" (desired id 3)
            'bbox': [200, 200, 50, 100],
            'area': 5000,
            'iscrowd': 0
        },
        {
            'id': 4,
            'image_id': 2,
            'category_id': 99,  # Not in desired_name2id, should be filtered out
            'bbox': [300, 300, 40, 40],
            'area': 1600,
            'iscrowd': 0
        },
    ],
    'categories': [
        {'id': 10, 'name': 'big_vehicle', 'supercategory': 'vehicle'},
        {'id': 11, 'name': 'car', 'supercategory': 'vehicle'},
        {'id': 12, 'name': 'human', 'supercategory': 'person'},
        {'id': 99, 'name': 'other', 'supercategory': 'other'},
    ]
}

ARGS = {
    'desired_name2id': desired_name2id,
    'coco_dict': coco_dict
}
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_slice_bboxes Not yet run
unverified (harvested, no recorded run)
obss/sahi/sahi/slicing.py
code served (permissive licence) · get_code("7fc4f6b132018cc2")
How it ran

no invocation recorded.

Also present in other papers' code

Of 19 samples lifted from this paper's linked repositories, 19 carry a function identity and 0 do not. Of the 19 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 19 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

Detection of small objects and objects far away in the scene is a major challenge in surveillance applications. Such objects are represented by small number of pixels in the image and lack sufficient details, making them difficult to detect using conventional detectors. In this work, an open-source framework called Slicing Aided Hyper Inference (SAHI) is proposed that provides a generic slicing aided inference and fine-tuning pipeline for small object detection. The proposed technique is generic in the sense that it can be applied on top of any available object detector without any fine-tuning. Experimental evaluations, using object detection baselines on the Visdrone and xView aerial object detection datasets show that the proposed inference method can increase object detection AP by 6.8%, 5.1% and 5.3% for FCOS, VFNet and TOOD detectors, respectively. Moreover, the detection accuracy can be further increased with a slicing aided fine-tuning, resulting in a cumulative increase of 12.7%, 13.4% and 14.5% AP in the same order. Proposed technique has been integrated with Detectron2, MMDetection and YOLOv5 models and it is publicly available at https://github.com/obss/sahi.git .

For agents

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

get_harvested_code_for_paper("2202.06934")
get_code_for_paper("2202.06934")
have("2202.06934")

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

curl https://syntology.ai/api/ran/2202.06934.json

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

[![Syntology run record](https://syntology.ai/api/ran/2202.06934.svg)](https://syntology.ai/paper/2202.06934)

Connect an agent — have() is free.