SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2404.13696 · 2024

Clio: Real-time Task-Driven Open-Set 3D Scene Graphs

arXiv · PDF · Open in the Atlas

Code that ran

15 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; 5 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 9 failures of our instrument; 2 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/2404.13696.json · badge: /api/ran/2404.13696.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
mit-spark/clio canonical (classifier label)
classifier basis: census
15 ran · 2 no recorded run
FunctionStatusWhere it lives
compute_cosine_sim Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
mit-spark/clio/clio_batch/helpers.py
code served (permissive licence) · get_code("0f04babc506ca3a1")
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 (1, 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 numpy as np
import copy

# Set up small test data
np.random.seed(0)

# Create v1: shape (2, 4) - 2 samples, 4-dimensional features
v1 = np.random.randn(2, 4)

# Create v2: shape (3, 4) - 3 tasks, 4-dimensional features
v2 = np.random.randn(3, 4)

# Create canonical_features: shape (3, 4) - 3 canonical features, 4-dimensional
canonical_features = np.random.randn(3, 4)

# Define arguments for compute_cosine_sim
ARGS = {
    "v1": v1,
    "v2": v2,
    "canonical_features": canonical_features,
    "use_lerf_canonical": True,
    "get_assignments": False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (1, 3) — shape (1, 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
compute_sim_to_tasks Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
mit-spark/clio/clio_batch/helpers.py
code served (permissive licence) · get_code("904d92e80020dd5b")
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, 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 numpy as np
import copy

# Set seed for reproducibility
np.random.seed(0)

# Create small test vectors with distinct dimensions
# v1: 2 samples, 4-dimensional features
v1 = np.random.randn(2, 4).astype(np.float32)

# v2: 3 samples, 4-dimensional features (different from v1 to show distinction)
v2 = np.random.randn(3, 4).astype(np.float32)

# canonical_features: 5 samples, 4-dimensional features (yet another size)
canonical_features = np.random.randn(5, 4).astype(np.float32)

ARGS = {
    "v1": v1,
    "v2": v2,
    "canonical_features": canonical_features,
    "use_lerf_canonical": True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (2, 3) — shape (2, 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
contains_object Ran
ran, violated the contract we drafted
mit-spark/clio/clio_batch/object_cluster.py
code served (permissive licence) · get_code("c20ac8b4610fedb0")
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 networkx as nx

class BoundingBox:
    """Stand-in bounding box class with is_inside method and min/max attributes."""
    def __init__(self, min_point, max_point):
        self.min = min_point
        self.max = max_point
    
    def is_inside(self, point):
        """Check if a point is inside this bounding box."""
        return all(self.min[i] <= point[i] <= self.max[i] for i in range(len(point)))

# Create a NetworkX directed graph
G_obj_nx = nx.DiGraph()

# Create bounding boxes for nodes
bbox1 = BoundingBox(min_point=(0.0, 0.0), max_point=(10.0, 10.0))
bbox2 = BoundingBox(min_point=(2.0, 2.0), max_point=(8.0, 8.0))

# Add nodes with bounding box attributes
G_obj_nx.add_node('node_1', bounding_box=bbox1)
G_obj_nx.add_node('node_2', bounding_box=bbox2)

# Define arguments
ARGS = {
    'node': 'node_1',
    'other_node': 'node_2',
    'G_obj_nx': G_obj_nx
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
networkx, 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
create_cluster_graphs Ran
ran; our contract draft was wrong, not the code
mit-spark/clio/clio_batch/object_cluster.py
code served (permissive licence) · get_code("cace94fae8881fde")
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 numpy as np
import networkx as nx

class BoundingBox:
    def __init__(self, min_coords, max_coords):
        self.min = min_coords
        self.max = max_coords

# Create a simple NetworkX graph with nodes and bounding boxes
G_obj_nx = nx.Graph()

# Add nodes with bounding box attributes
bbox_1 = BoundingBox(
    min_coords=np.array([0.0, 0.0]),
    max_coords=np.array([2.0, 2.0])
)
bbox_2 = BoundingBox(
    min_coords=np.array([1.5, 1.5]),
    max_coords=np.array([4.0, 4.0])
)
bbox_3 = BoundingBox(
    min_coords=np.array([10.0, 10.0]),
    max_coords=np.array([12.0, 12.0])
)

G_obj_nx.add_node("node_1", bounding_box=bbox_1)
G_obj_nx.add_node("node_2", bounding_box=bbox_2)
G_obj_nx.add_node("node_3", bounding_box=bbox_3)

# Create partition for testing
partition = [{"node_1", "node_2"}, {"node_3"}]

# Bounding box dilation factor
bbox_dilation = 0.1

ARGS = {
    "G_obj_nx": G_obj_nx,
    "partition": partition,
    "bbox_dilation": bbox_dilation
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
networkx, numpy — 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
create_obb_from_dict Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
mit-spark/clio/clio_data_maker/rescale_bbox.py
code served (permissive licence) · get_code("0afd6939ef1f5ea5")
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 numpy as np
from scipy.spatial.transform import Rotation as R

# Create a sample bbox_dict with rotation as quaternion and center coordinates
bbox_dict = {
    'rotation': {
        'w': 1.0,
        'x': 0.0,
        'y': 0.0,
        'z': 0.0
    },
    'center': [1.0, 2.0, 3.0],
    'extents': [2.0, 3.0, 4.0]
}

# Create a 4x4 transformation matrix
transform_mat = np.array([
    [1.0, 0.0, 0.0, 0.5],
    [0.0, 1.0, 0.0, 1.5],
    [0.0, 0.0, 1.0, 2.5],
    [0.0, 0.0, 0.0, 1.0]
], dtype=np.float64)

ARGS = {
    'bbox_dict': bbox_dict,
    'transform_mat': transform_mat
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — 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
default_device Ran
ran; our contract draft was wrong, not the code
mit-spark/clio/clio_batch/helpers.py
code served (permissive licence) · get_code("81cc50e1fa115ce7")
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. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type 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)

ARGS = {
    'use_cuda': False
}
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
get_box_verticies Ran
ran, honoured the contract we drafted
mit-spark/clio/clio_eval/iou.py
code served (permissive licence) · get_code("b18d5a985faa6004")
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 (9, 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 numpy as np

class MockOBB:
    def __init__(self):
        self.center_point = np.array([1.0, 2.0, 3.0])
        self.box_pts = np.array([
            [0.0, 0.0, 0.0],
            [1.0, 1.0, 1.0],
            [2.0, 2.0, 2.0],
            [3.0, 3.0, 3.0],
            [4.0, 4.0, 4.0],
            [5.0, 5.0, 5.0],
            [6.0, 6.0, 6.0],
            [7.0, 7.0, 7.0],
        ])
    
    def get_center(self):
        return self.center_point
    
    def get_box_points(self):
        return self.box_pts

obb = MockOBB()

ARGS = {
    'obb': obb
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (9, 3) — shape (9, 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
get_dsg_version Ran
ran; our contract draft was wrong, not the code
mit-spark/clio/clio_eval/evaluate_helpers.py
code served (permissive licence) · get_code("89c8f9177bf25da3")
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 JSON file with DSG structure
temp_dir = tempfile.mkdtemp()
dsg_file = os.path.join(temp_dir, "test_dsg.json")

# Create a valid DSG file with SPARK_DSG_header and version info
dsg_data = {
    "SPARK_DSG_header": {
        "version": {
            "major": 1,
            "minor": 2,
            "patch": 3
        }
    },
    "other_data": "some_value"
}

with open(dsg_file, 'w') as f:
    json.dump(dsg_data, f)

ARGS = {
    "dsg_file": dsg_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
get_last_feature Ran
ran, honoured the contract we drafted
mit-spark/clio/clio_batch/cluster_utilities.py
code served (permissive licence) · get_code("308809ffeae78b00")
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 1, dtype family float, shape (2,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Test case 1: 2D array (typical case)
features_2d = np.array([[1.0, 2.0, 3.0, 4.0],
                        [5.0, 6.0, 7.0, 8.0]])

# Test case 2: 1D array (edge case that returns as-is)
features_1d = np.array([1.0, 2.0, 3.0, 4.0])

# Using the 2D case as the primary test
ARGS = {
    "features": features_2d
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (2,) — shape (2,) 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
is_point_inside_box Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
mit-spark/clio/clio_eval/utils.py
code served (permissive licence) · get_code("85f1beae457883d3")
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

torch.manual_seed(0)

# Construct arguments for is_point_inside_box
# The function checks if a 3D point is within a bounding box defined by 6 bounds values

# point: a 3-element sequence representing (x, y, z) coordinates
point = [5.0, 5.0, 5.0]

# bounds: a 6-element sequence representing [x_min, y_min, z_min, x_max, y_max, z_max]
bounds = [0.0, 0.0, 0.0, 10.0, 10.0, 10.0]

ARGS = {
    "point": point,
    "bounds": bounds
}
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 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
max_from_box_position Ran
ran, honoured the contract we drafted
mit-spark/clio/clio_eval/utils.py
code served (permissive licence) · get_code("2292882fb71b75af")
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 float, shape (2, 1, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

class BoundingBox:
    def __init__(self):
        self.dimensions = np.array([[1.0, 2.0, 3.0, 4.0],
                                     [5.0, 6.0, 7.0, 8.0]])

class Node:
    def __init__(self):
        self.bounding_box = BoundingBox()
        self.position = np.array([[10.0, 20.0, 30.0, 40.0],
                                  [50.0, 60.0, 70.0, 80.0]])

node = Node()

ARGS = {
    "node": node,
    "offset_to_lower_corner": False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family float, shape (2, 1, 4) — shape (2, 1, 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
min_from_box_position Ran
ran, honoured the contract we drafted
mit-spark/clio/clio_eval/utils.py
code served (permissive licence) · get_code("1c3914ffaf0ba4d0")
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 float, shape (4, 1, 1) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

class BoundingBox:
    def __init__(self, dimensions):
        self.dimensions = dimensions

class Node:
    def __init__(self, position, bounding_box):
        self.position = position
        self.bounding_box = bounding_box

# Create position array with shape (4, 1) - 4 spatial dimensions
position = np.array([[1.0], [2.0], [3.0], [4.0]])

# Create bounding box dimensions with shape (4, 1) - 4 spatial dimensions
dimensions = np.array([[0.5], [0.6], [0.7], [0.8]])

# Create bounding box object
bbox = BoundingBox(dimensions=dimensions)

# Create node with position and bounding box
node = Node(position=position, bounding_box=bbox)

ARGS = {
    "node": node,
    "offset_to_lower_corner": False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family float, shape (4, 1, 1) — shape (4, 1, 1) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
mutual_information Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
mit-spark/clio/clio_batch/information_metrics.py
code served (permissive licence) · get_code("109cd6e7b1ebd161")
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 float64, ndim 0, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create valid probability mass functions (PMFs)
# px: probability distribution over X (1D array)
# pc: probability distribution over C (1D array)
# pc_x: conditional probability P(C|X) (2D array where rows are C, cols are X)

np.random.seed(0)

# Create px: marginal probability over X, shape (4,)
px = np.array([0.1, 0.3, 0.4, 0.2])

# Create pc: marginal probability over C, shape (3,)
pc = np.array([0.2, 0.5, 0.3])

# Create pc_x: conditional probability P(C|X), shape (3, 4)
# Each column should sum to 1 (probability over C given each X)
pc_x = np.array([
    [0.5, 0.2, 0.1, 0.4],
    [0.3, 0.6, 0.7, 0.4],
    [0.2, 0.2, 0.2, 0.2]
])

ARGS = {
    "px": px,
    "pc": pc,
    "pc_x": pc_x,
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
rescale Ran
ran; our contract draft was wrong, not the code
mit-spark/clio/clio_data_maker/rescale_bbox.py
code served (permissive licence) · get_code("1f9d9ed8032cec88")
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
import numpy as np

torch.manual_seed(0)

# Analyze the function: it takes a detection dict with 'center', 'extents', and 'rotation' keys
# and a scale parameter. It multiplies center and extents by scale, keeps rotation unchanged.

detection = {
    'center': [1.0, 2.0, 3.0],
    'extents': [0.5, 1.5, 2.5],
    'rotation': 45.0
}

scale = 2.0

ARGS = {
    'detection': detection,
    'scale': scale
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
shannon_entropy Ran
ran, honoured the contract we drafted
mit-spark/clio/clio_batch/information_metrics.py
code served (permissive licence) · get_code("31fec44a12e1775a")
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 1, dtype family float, shape (2,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create a valid probability mass function (PMF)
# Shannon entropy expects probabilities that sum to 1
# Create a 2D array where each column is a valid PMF
p = np.array([
    [0.25, 0.5],
    [0.25, 0.3],
    [0.25, 0.15],
    [0.25, 0.05]
], dtype=np.float64)

# Verify each column sums to 1 (valid PMF)
assert np.allclose(np.sum(p, axis=0), 1.0), "Columns must sum to 1"

# Ensure no zeros that would cause log(0) issues
p = np.clip(p, 1e-10, 1.0)

ARGS = {
    "p": p
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (2,) — shape (2,) 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
bboxes_intersect Not yet run
unverified (harvested, no recorded run)
mit-spark/clio/clio_batch/object_cluster.py
code served (permissive licence) · get_code("958a2e2335058ed4")
How it ran

no invocation recorded.

js_divergence Not yet run
unverified (harvested, no recorded run)
mit-spark/clio/clio_batch/information_metrics.py
code served (permissive licence) · get_code("81689c2e5f8bb283")
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

Modern tools for class-agnostic image segmentation (e.g., SegmentAnything) and open-set semantic understanding (e.g., CLIP) provide unprecedented opportunities for robot perception and mapping. While traditional closed-set metric-semantic maps were restricted to tens or hundreds of semantic classes, we can now build maps with a plethora of objects and countless semantic variations. This leaves us with a fundamental question: what is the right granularity for the objects (and, more generally, for the semantic concepts) the robot has to include in its map representation? While related work implicitly chooses a level of granularity by tuning thresholds for object detection, we argue that such a choice is intrinsically task-dependent. The first contribution of this paper is to propose a task-driven 3D scene understanding problem, where the robot is given a list of tasks in natural language and has to select the granularity and the subset of objects and scene structure to retain in its map that is sufficient to complete the tasks. We show that this problem can be naturally formulated using the Information Bottleneck (IB), an established information-theoretic framework. The second contribution is an algorithm for task-driven 3D scene understanding based on an Agglomerative IB approach, that is able to cluster 3D primitives in the environment into task-relevant objects and regions and executes incrementally. The third contribution is to integrate our task-driven clustering algorithm into a real-time pipeline, named Clio, that constructs a hierarchical 3D scene graph of the environment online using only onboard compute, as the robot explores it. Our final contribution is an extensive experimental campaign showing that Clio not only allows real-time construction of compact open-set 3D scene graphs, but also improves the accuracy of task execution by limiting the map to relevant semantic concepts.

For agents

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

get_harvested_code_for_paper("2404.13696")
get_code_for_paper("2404.13696")
have("2404.13696")

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

curl https://syntology.ai/api/ran/2404.13696.json

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

[![Syntology run record](https://syntology.ai/api/ran/2404.13696.svg)](https://syntology.ai/paper/2404.13696)

Connect an agent — have() is free.