SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2203.05238 · CVPR · 2022

Back to Reality: Weakly-supervised 3D Object Detection with Shape-guided Label Enhancement

Yu Zheng, Yongming Rao, Jiwen Lu, Jie Zhou, Yifan Wang, Xiuwei Xu

arXiv · PDF · Open in the Atlas

Code that ran

20 of 25 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 13 have no contract verdict; 5 failures of our instrument; 5 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/2203.05238.json · badge: /api/ran/2203.05238.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
wyf-ACCEPT/BackToReality canonical (classifier label)
classifier basis not recorded on the link
13 ran · 1 no recorded run
xuxw98/backtoreality marked official in Papers with Code
Papers with Code archive
7 ran · 4 no recorded run
FunctionStatusWhere it lives
Is_horizontal Ran
ran on a synthesized input
wyf-ACCEPT/BackToReality/data_generation/ScanNet/segment_tools.py
code served (permissive licence) · get_code("af68a34722851d0e")
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 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)

# The function accesses scene_vertices[seg] to get a subset of vertices,
# then accesses the third column ([:, 2]) to get z-coordinates.
# It calls .sort() on the result and .shape[0] to get the number of elements.
# seg should be an indexable collection that selects rows from scene_vertices.

# Create scene_vertices as a tensor with shape (10, 3) - 10 vertices, 3D coordinates
scene_vertices = torch.randn(10, 3)

# Create seg as a tensor of indices to select from scene_vertices
# This will select 6 vertices from the 10 available
seg = torch.tensor([0, 1, 3, 5, 7, 9])

ARGS = {
    "scene_vertices": scene_vertices,
    "seg": seg
}
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 Tensor, 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
SigmoidFocalClassificationLoss Ran
ran on a synthesized input
xuxw98/backtoreality/detection/GroupFree3D/models/loss_helper.py
code served (permissive licence) · get_code("2616a152109ab293")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, 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

class SigmoidFocalClassificationLoss(nn.Module):
    """
    Sigmoid focal cross entropy loss.
    """

    def __init__(self, gamma: float=2.0, alpha: float=0.25):
        """
        Args:
            gamma: Weighting parameter to balance loss for hard and easy examples.
            alpha: Weighting parameter to balance loss for positive and negative examples.
        """
        super(SigmoidFocalClassificationLoss, self).__init__()
        self.alpha = alpha
        self.gamma = gamma

    @staticmethod
    def sigmoid_cross_entropy_with_logits(input: torch.Tensor, target: torch.Tensor):
        """ PyTorch Implementation for tf.nn.sigmoid_cross_entropy_with_logits:
            max(x, 0) - x * z + log(1 + exp(-abs(x))) in
            https://www.tensorflow.org/api_docs/python/tf/nn/sigmoid_cross_entropy_with_logits

        Args:
            input: (B, #proposals, #classes) float tensor.
                Predicted logits for each class
            target: (B, #proposals, #classes) float tensor.
                One-hot encoded classification targets

        Returns:
            loss: (B, #proposals, #classes) float tensor.
                Sigmoid cross entropy loss without reduction
        """
        loss = torch.clamp(input, min=0) - input * target + torch.log1p(torch.exp(-torch.abs(input)))
        return loss

    def forward(self, input: torch.Tensor, target: torch.Tensor, weights: torch.Tensor):
        """
        Args:
            input: (B, #proposals, #classes) float tensor.
                Predicted logits for each class
            target: (B, #proposals, #classes) float tensor.
                One-hot encoded classification targets
            weights: (B, #proposals) float tensor.
                Anchor-wise weights.

        Returns:
            weighted_loss: (B, #proposals, #classes) float tensor after weighting.
        """
        pred_sigmoid = torch.sigmoid(input)
        alpha_weight = target * self.alpha + (1 - target) * (1 - self.alpha)
        pt = target * (1.0 - pred_sigmoid) + (1.0 - target) * pred_sigmoid
        focal_weight = alpha_weight * torch.pow(pt, self.gamma)
        bce_loss = self.sigmoid_cross_entropy_with_logits(input, target)
        loss = focal_weight * bce_loss
        weights = weights.unsqueeze(-1)
        assert weights.shape.__len__() == loss.shape.__len__()
        return loss * weights


INIT_ARGS = {
    'gamma': 2.0,
    'alpha': 0.25,
}

ARGS = {
    'input': torch.randn(2, 5, 4),
    'target': torch.randint(0, 2, (2, 5, 4)).float(),
    'weights': torch.ones(2, 5),
}

if __name__ == '__main__':
    loss_fn = SigmoidFocalClassificationLoss(**INIT_ARGS)
    output = loss_fn(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output:\n{output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
compute_objectness_loss_based_on_query_points Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
xuxw98/backtoreality/detection/GroupFree3D/models/loss_helper.py
code served (permissive licence) · get_code("642ad5f6031a8aa1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. 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 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)

# Create small tensors with distinct dimensions
B = 2  # batch size
N = 4  # number of points
K = 3  # number of query points
K2 = 5  # number of ground truth objects
num_classes = 1

# Build end_points dict with all required keys
end_points = {
    'seed_inds': torch.randint(0, N, (B, N)),
    'center_label': torch.randn(B, K2, 3),
    'query_points_sample_inds': torch.randint(0, N, (B, K)),
    'point_obj_mask': torch.randint(0, 2, (B, N)).float(),
    'point_instance_label': torch.randint(-1, K2, (B, N)),
    'proposal_objectness_scores': torch.randn(B, K, num_classes),
}

# Create additional objectness_scores for decoder layers if needed
num_decoder_layers = 0

ARGS = {
    'end_points': end_points,
    'num_decoder_layers': num_decoder_layers,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
compute_points_obj_cls_loss_hard_topk Ran
ran, honoured the contract we drafted
xuxw98/backtoreality/detection/GroupFree3D/models/loss_helper.py
code served (permissive licence) · get_code("c9e3c6528cd69995")
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 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 torch.nn as nn

torch.manual_seed(0)

# Build minimal end_points dict with all required keys
B = 2  # batch size
K = 4  # number of seeds/proposals
K2 = 3  # number of ground truth objects
num_classes = 1

end_points = {
    'box_label_mask': torch.ones((B, K2), dtype=torch.float32),
    'seed_inds': torch.randint(0, K, (B, K), dtype=torch.long),
    'seed_xyz': torch.randn((B, K, 3), dtype=torch.float32),
    'seeds_obj_cls_logits': torch.randn((B, K, 1), dtype=torch.float32),
    'center_label': torch.randn((B, K2, 3), dtype=torch.float32),
    'size_gts': torch.randn((B, K2, 3), dtype=torch.float32),
    'point_instance_label': torch.randint(-1, K2, (B, K), dtype=torch.long),
}

topk = 2

ARGS = {
    'end_points': end_points,
    'topk': topk,
}
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 Tensor, 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
compute_points_obj_cls_loss_hard_topk_weak Ran
ran, honoured the contract we drafted
xuxw98/backtoreality/detection/GroupFree3D/models/loss_helper.py
code served (permissive licence) · get_code("5d501aaa66769592")
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 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 torch.nn as nn

torch.manual_seed(0)

# Build end_points dict with all required keys
B = 2  # batch size
K = 4  # number of seeds
K2 = 3  # number of gt boxes
num_classes = 1

end_points = {
    'box_label_mask': torch.ones((B, K2), dtype=torch.float32),
    'seed_inds': torch.randint(0, K, (B, K), dtype=torch.long),
    'seed_xyz': torch.randn((B, K, 3), dtype=torch.float32),
    'seeds_obj_cls_logits': torch.randn((B, K, num_classes), dtype=torch.float32),
    'center_label': torch.randn((B, K2, 3), dtype=torch.float32),
}

topk = 2

ARGS = {
    'end_points': end_points,
    'topk': topk,
}
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 Tensor, 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
find_nearest_object Ran
ran on a synthesized input
wyf-ACCEPT/BackToReality/data_generation/ScanNet/scannet_scene_synthesis.py
code served (permissive licence) · get_code("cc8c7c54286373af")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

# Construct the info_dict parameter
# Structure: key -> [[[numerator, denominator], ...], ..., support_flag]
# The function accesses value[0][0][0], value[0][0][1], and value[2]
info_dict = {
    'obj_code_1': [
        [[8, 4], [2, 1]],  # value[0] is list of lists
        'some_metadata',
        True  # value[2] is the support flag
    ],
    'obj_code_2': [
        [[10, 5], [3, 2]],
        'other_metadata',
        False
    ],
    'obj_code_3': [
        [[6, 3], [1, 2]],
        'more_metadata',
        True
    ],
    'obj_code_4': [
        [[0, 0], [0, 0]],  # Division by zero case - should be skipped
        'bad_metadata',
        True
    ]
}

# ls_ratio: a float representing the target width-to-height ratio
ls_ratio = 2.0

# object_name: a string prefix for the output filename
object_name = 'test_object'

# require_support: boolean flag to filter by support
require_support = False

ARGS = {
    'ls_ratio': ls_ratio,
    'info_dict': info_dict,
    'object_name': object_name,
    'require_support': require_support
}
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 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
get_MER Ran
ran on a synthesized input
wyf-ACCEPT/BackToReality/data_generation/Matterport3D/modelnet40_tools.py
code served (permissive licence) · get_code("5909c81367fe7e2b")
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
import cv2

class MockPointCloud:
    def __init__(self, points):
        self.points = points

# Create a simple point cloud with 5 points in 2D (we only use first 2 coords)
# Points should form a reasonable shape for minAreaRect
points_data = np.array([
    [0.0, 0.0, 0.5],
    [1.0, 0.0, 0.5],
    [1.0, 0.5, 0.5],
    [0.0, 0.5, 0.5],
    [0.5, 0.25, 0.5]
], dtype=np.float32)

pcd = MockPointCloud(points_data)

ARGS = {
    'pcd': pcd
}
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 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
get_MER Ran
ran on a synthesized input
wyf-ACCEPT/BackToReality/data_generation/Matterport3D/matterport_scene_synthesis.py
code served (permissive licence) · get_code("f66ccab4c6f93ada")
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
import cv2
import copy

# Create a small set of 2D points as a numpy array
# Shape should be (N, 2+) where first 2 columns are x, y coordinates
points = np.array([
    [0.5, 0.3],
    [0.7, 0.2],
    [0.8, 0.5],
    [0.6, 0.7]
], dtype=np.float32)

ARGS = {
    'points': points
}
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 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
get_id_to_label Ran
ran on a synthesized input
wyf-ACCEPT/BackToReality/data_generation/ScanNet/ins_ply.py
code served (permissive licence) · get_code("114d8fa311c666a5")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import json
import tempfile

# Create a temporary JSON file with the expected structure
temp_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.json')
temp_filename = temp_file.name

# Write sample data matching the function's expected structure
sample_data = {
    'segGroups': [
        {
            'objectId': 0,
            'label': 'chair'
        },
        {
            'objectId': 1,
            'label': 'table'
        },
        {
            'objectId': 2,
            'label': 'wall'
        }
    ]
}

json.dump(sample_data, temp_file)
temp_file.close()

ARGS = {
    'filename': temp_filename
}
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_normal Ran
ran on a synthesized input
wyf-ACCEPT/BackToReality/data_generation/ScanNet/segment_tools.py
code served (permissive licence) · get_code("52dc1aebbd2b533c")
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 (4, 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 MockMesh:
    def __init__(self):
        # Create a simple tetrahedron mesh with 4 vertices and 4 triangular faces
        self.vertices = np.array([
            [0.0, 0.0, 0.0],
            [1.0, 0.0, 0.0],
            [0.5, 1.0, 0.0],
            [0.5, 0.5, 1.0]
        ], dtype=np.float64)
        
        # Define 4 triangular faces (indices into vertices)
        self.faces = np.array([
            [0, 1, 2],
            [0, 1, 3],
            [1, 2, 3],
            [0, 2, 3]
        ], dtype=np.int32)
        
        self.triangle_normals = None
    
    def compute_vertex_normals(self):
        # Compute triangle normals from vertices and faces
        triangle_normals = []
        for face in self.faces:
            v0 = self.vertices[face[0]]
            v1 = self.vertices[face[1]]
            v2 = self.vertices[face[2]]
            
            # Compute two edge vectors
            edge1 = v1 - v0
            edge2 = v2 - v0
            
            # Compute normal via cross product
            normal = np.cross(edge1, edge2)
            norm = np.linalg.norm(normal)
            if norm > 0:
                normal = normal / norm
            
            triangle_normals.append(normal)
        
        self.triangle_normals = np.array(triangle_normals)

mesh = MockMesh()

ARGS = {
    "mesh": mesh
}
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 (4, 3) — shape (4, 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_solid_MER Ran
ran on a synthesized input
wyf-ACCEPT/BackToReality/data_generation/ScanNet/scannet_scene_synthesis.py
code served (permissive licence) · get_code("22667864de7fcab8")
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
import torch

torch.manual_seed(0)

# The function takes a single parameter 'points'
# Looking at the function body:
# - points[:, 0:2] extracts first 2 columns
# - points is multiplied by 1000 and cast to int
# - points is used with cv2.minAreaRect which expects a 2D array of point coordinates
# - points is used with KMeans.predict which expects a 2D array
# So points should be a 2D numpy array with at least 2 columns

# Create a small sample of points with shape (n_points, at_least_2_columns)
# Using shape (6, 3) to have some distinct dimension from other possible uses
points = np.array([
    [0.1, 0.2, 0.5],
    [0.15, 0.25, 0.5],
    [0.12, 0.22, 0.5],
    [0.5, 0.6, 0.5],
    [0.55, 0.65, 0.5],
    [0.52, 0.62, 0.5]
], dtype=np.float64)

ARGS = {
    "points": points
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy, sklearn, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_solid_MER Ran
ran on a synthesized input
wyf-ACCEPT/BackToReality/data_generation/Matterport3D/matterport_scene_synthesis.py
code served (permissive licence) · get_code("d7006f63008a044b")
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
import cv2
import copy
from sklearn.cluster import KMeans

# Create a simple set of points for testing
# The function expects points as a numpy array with at least columns 0:2 (x, y coordinates)
# Shape should be (num_points, num_features) where num_features >= 2
points = np.array([
    [0.5, 0.5, 0.1, 0.2],
    [0.6, 0.4, 0.15, 0.25],
    [0.7, 0.6, 0.12, 0.22],
    [1.2, 1.3, 0.18, 0.28],
    [1.3, 1.4, 0.14, 0.24],
    [1.4, 1.2, 0.16, 0.26],
], dtype=np.float32)

ARGS = {
    "points": points,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy, sklearn — 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
get_support_z Ran
ran on a synthesized input
wyf-ACCEPT/BackToReality/data_generation/Matterport3D/modelnet40_tools.py
code served (permissive licence) · get_code("d9e542fd2784b3d8")
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 float32, 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 numpy as np

class MockPointCloud:
    def __init__(self, points, normals):
        self.points = points
        self.normals = normals

# Create test data
# 20 points with (x, y, z) coordinates
np.random.seed(0)
points = np.random.randn(20, 3).astype(np.float32)

# Create normals where some have high z-component (> 0.88)
normals = np.random.randn(20, 3).astype(np.float32)
# Normalize normals
normals = normals / np.linalg.norm(normals, axis=1, keepdims=True)

# Ensure some normals have abs(dot(normal, [0,0,1])) > 0.88
# This means the z-component should be > 0.88 or < -0.88
for i in range(10):
    normals[i] = np.array([0.0, 0.0, 0.95])
for i in range(10, 15):
    normals[i] = np.array([0.0, 0.0, -0.95])

# Create point cloud object
pcd = MockPointCloud(points, normals)

ARGS = {
    "pcd": pcd
}
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 float32, 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
l1_loss Ran
ran; our contract draft was wrong, not the code
xuxw98/backtoreality/detection/GroupFree3D/models/loss_helper.py
code served (permissive licence) · get_code("e1e4a142a595bcf2")
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, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create a small error tensor (batch size 2, dimension 4)
error = torch.randn(2, 4)

ARGS = {
    "error": error
}
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 Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 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
modelnet40_to_id Ran
ran on a synthesized input
wyf-ACCEPT/BackToReality/data_generation/ScanNet/scannet_utils.py
code served (permissive licence) · get_code("fde3fc16b0a466d2")
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 int, ndim 0, dtype family int, 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 tempfile
import os

# Create a temporary modelnet40_shape_names.txt file
temp_dir = tempfile.gettempdir()
shape_names_path = os.path.join(temp_dir, 'modelnet40_shape_names.txt')

# Write sample ModelNet40 shape names to the file
shape_names = [
    'airplane',
    'bathtub',
    'bed',
    'bench',
    'bicycle',
    'birdhouse',
    'bookshelf',
    'bottle',
    'bowl',
    'bus',
    'cabinet',
    'camera',
    'can',
    'cap',
    'car',
    'chair',
    'clipart',
    'clock',
    'computer',
    'cone',
    'curtain',
    'desk',
    'dishwasher',
    'display',
    'door',
    'dresser',
    'dumbbell',
    'earphone',
    'fan',
    'file',
    'flower_pot',
    'fountain',
    'glass_box',
    'guitar',
    'gun',
    'gym_shoes',
    'hair_brush',
    'hat',
    'helmet',
    'home',
    'house'
]

with open(shape_names_path, 'w') as f:
    for name in shape_names:
        f.write(name + '\n')

# Change to the temp directory so the function can find the file
original_cwd = os.getcwd()
os.chdir(temp_dir)

ARGS = {
    'modelname': 'chair'
}
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 int, ndim 0, dtype family int, 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
nn_distance Ran
ran; our contract draft was wrong, not the code
xuxw98/backtoreality/detection/GroupFree3D/models/loss_helper.py
code served (permissive licence) · get_code("60f5b7fe1cc54f17")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

def huber_loss(x, delta):
    """Stand-in for huber_loss since it's not defined in the function"""
    mask = torch.abs(x) <= delta
    return torch.where(mask, 0.5 * x ** 2, delta * (torch.abs(x) - 0.5 * delta))

# Create small test tensors
# pc1: (B=2, N=3, C=4)
pc1 = torch.randn(2, 3, 4)

# pc2: (B=2, M=5, C=4)
pc2 = torch.randn(2, 5, 4)

ARGS = {
    "pc1": pc1,
    "pc2": pc2,
    "l1smooth": False,
    "delta": 1.0,
    "l1": 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 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
read_aggregation Ran
ran on a synthesized input
wyf-ACCEPT/BackToReality/data_generation/ScanNet/ins_ply.py
code served (permissive licence) · get_code("99136f050d5a159b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import json
import tempfile

# Create a temporary JSON file with valid aggregation data
temp_file = tempfile.NamedTemporaryFile(mode='w', suffix='.json', delete=False)
aggregation_data = {
    'segGroups': [
        {
            'objectId': 0,
            'label': 'wall',
            'segments': [0, 1, 2]
        },
        {
            'objectId': 1,
            'label': 'floor',
            'segments': [3, 4]
        },
        {
            'objectId': 2,
            'label': 'wall',
            'segments': [5]
        }
    ]
}
json.dump(aggregation_data, temp_file)
temp_file.close()

ARGS = {
    'filename': temp_file.name
}
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, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
read_label_mapping Ran
ran on a synthesized input
wyf-ACCEPT/BackToReality/data_generation/ScanNet/scannet_utils.py
code served (permissive licence) · get_code("69e7d6553f196430")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import csv
import tempfile

# Create a temporary directory and files for testing
tmpdir = tempfile.mkdtemp()

# Create modelnet40_shape_names.txt file
modelnet40_path = os.path.join(tmpdir, 'modelnet40_shape_names.txt')
with open(modelnet40_path, 'w') as f:
    f.write('airplane\n')
    f.write('bathtub\n')
    f.write('bed\n')
    f.write('bench\n')
    f.write('bicycle\n')

# Create a label mapping CSV file
mapping_path = os.path.join(tmpdir, 'label_mapping.tsv')
with open(mapping_path, 'w') as f:
    f.write('category\tModelNet40\n')
    f.write('syn_airplane\tairplane\n')
    f.write('syn_bathtub\tbathtub\n')
    f.write('syn_bed\tbed\n')

# Change to the temp directory so the function can find modelnet40_shape_names.txt
original_cwd = os.getcwd()
os.chdir(tmpdir)

# Now define the arguments
ARGS = {
    'filename': mapping_path,
    'label_from': 'category',
    'label_to': 'ModelNet40'
}
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
read_segmentation Ran
ran; our contract draft was wrong, not the code
wyf-ACCEPT/BackToReality/data_generation/ScanNet/ins_ply.py
code served (permissive licence) · get_code("527894e637039fb3")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import json
import tempfile

# Create a temporary JSON file with valid segmentation data
temp_dir = tempfile.mkdtemp()
temp_filename = os.path.join(temp_dir, "test_segmentation.json")

# Create sample segmentation data
segmentation_data = {
    "segIndices": [0, 1, 0, 2, 1, 2, 0, 1]
}

with open(temp_filename, 'w') as f:
    json.dump(segmentation_data, f)

ARGS = {
    "filename": temp_filename
}
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, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
smoothl1_loss Ran
ran; our contract draft was wrong, not the code
xuxw98/backtoreality/detection/GroupFree3D/models/loss_helper.py
code served (permissive licence) · get_code("a18fa7b4ad55d938")
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, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create a small error tensor (batch size 2, dimension 4)
error = torch.randn(2, 4)

# Define delta parameter
delta = 1.0

# Construct the ARGS dict with all parameters
ARGS = {
    "error": error,
    "delta": delta,
}
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 Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 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
compute_box_and_sem_cls_loss Not yet run
unverified (harvested, no recorded run)
xuxw98/backtoreality/detection/GroupFree3D/models/loss_helper.py
code served (permissive licence) · get_code("7d9d44a980e24c22")
How it ran

no invocation recorded.

compute_center_and_sem_cls_loss Not yet run
unverified (harvested, no recorded run)
xuxw98/backtoreality/detection/GroupFree3D/models/loss_helper.py
code served (permissive licence) · get_code("4a436f36baabf7ee")
How it ran

no invocation recorded.

compute_objectness_loss_based_on_query_points_weak Not yet run
unverified (harvested, no recorded run)
xuxw98/backtoreality/detection/GroupFree3D/models/loss_helper.py
code served (permissive licence) · get_code("137a1dae05a132ca")
How it ran

no invocation recorded.

get_loss_weak Not yet run
unverified (harvested, no recorded run)
xuxw98/backtoreality/detection/GroupFree3D/models/loss_helper.py
code served (permissive licence) · get_code("a8c59f9640373662")
How it ran

no invocation recorded.

read_mesh_vertices Not yet run
unverified (harvested, no recorded run)
wyf-ACCEPT/BackToReality/data_generation/ScanNet/scannet_utils.py
code served (permissive licence) · get_code("f772d38e5bc06661")
How it ran

no invocation recorded.

Also present in other papers' code

Of 25 samples lifted from this paper's linked repositories, 25 carry a function identity and 0 do not. Of the 25 distinct functions with an identity, 4 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 1 the same month, 4 later) and 21 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
read_mesh_vertices 5 other papers
1 the same month, 4 later
2203.09065 (2022-03)
STPLS3D: A Large-Scale Synthetic and Real Aerial Photogrammetry 3D Point Cloud Dataset
aaai_25335 (2023 (year only))
Superpoint Transformer for 3D Scene Instance Segmentation
2303.00246 (2023-03)
ISBNet: a 3D Point Cloud Instance Segmentation Network with Instance-aware Sampling and Box-aware Dynamic Convolution
and 2 more not named here
l1_loss
brick of this paper's assembly: helper
2 other papers
2 later
2204.06272 (2022-04)
3D-SPS: Single-Stage 3D Visual Grounding via Referred Point Progressive Selection
2402.15933 (2024-02)
Bridging the Gap between 2D and 3D Visual Question Answering: A Fusion Approach for 3D VQA
smoothl1_loss
brick of this paper's assembly: helper
2 other papers
2 later
2204.06272 (2022-04)
3D-SPS: Single-Stage 3D Visual Grounding via Referred Point Progressive Selection
2402.15933 (2024-02)
Bridging the Gap between 2D and 3D Visual Question Answering: A Fusion Approach for 3D VQA
read_segmentation 1 other paper
1 later
2403.04198 (2024-03)
This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore

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

In this paper, we propose a weakly-supervised approach for 3D object detection, which makes it possible to train a strong 3D detector with position-level annotations (i.e. annotations of object centers). In order to remedy the information loss from box annotations to centers, our method, namely Back to Reality (BR), makes use of synthetic 3D shapes to convert the weak labels into fully-annotated virtual scenes as stronger supervision, and in turn utilizes the perfect virtual labels to complement and refine the real labels. Specifically, we first assemble 3D shapes into physically reasonable virtual scenes according to the coarse scene layout extracted from position-level annotations. Then we go back to reality by applying a virtualto-real domain adaptation method, which refine the weak labels and additionally supervise the training of detector with the virtual scenes. Furthermore, we propose a more challenging benckmark for indoor 3D object detection with more diversity in object sizes for better evaluation. With less than 5% of the labeling labor, we achieve comparable detection performance with some popular fully-supervised approaches on the widely used ScanNet dataset. Code is available at: https://github.com/wyf-ACCEPT/BackToReality.

For agents

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

get_harvested_code_for_paper("2203.05238")
get_code_for_paper("2203.05238")
have("2203.05238")

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

curl https://syntology.ai/api/ran/2203.05238.json

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

[![Syntology run record](https://syntology.ai/api/ran/2203.05238.svg)](https://syntology.ai/paper/2203.05238)

Connect an agent — have() is free.