SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1905.10296 · 2019

Uncertainty Estimation in One-Stage Object Detection

arXiv · PDF · Open in the Atlas

Code that ran

11 of 13 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 9 have no contract verdict; 2 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/1905.10296.json · badge: /api/ran/1905.10296.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
flkraus/bayesian-yolov3 canonical (classifier label)
classifier basis: census
11 ran · 2 no recorded run
FunctionStatusWhere it lives
bbox_to_ecp_format Ran
ran on a synthesized input
flkraus/bayesian-yolov3/inference_standard_yolov3.py
code served (permissive licence) · get_code("24e74d320a0d45b8")
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
import torch
import torch.nn as nn

# Create a minimal model that has the attributes accessed in the function
class MinimalModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.cls_start_idx = 5
        self.cls_cnt = 2
        self.obj_idx = 4

model = MinimalModel()

# Create bbox as a numpy array with values in appropriate ranges
# bbox needs at least: indices 0-3 (coordinates), obj_idx (4), cls_start_idx to cls_start_idx+cls_cnt (5-6)
bbox = np.array([
    0.1,      # bbox[0] - normalized y0
    0.2,      # bbox[1] - normalized x0
    0.8,      # bbox[2] - normalized y1
    0.9,      # bbox[3] - normalized x1
    0.95,     # bbox[4] - objectness score at obj_idx
    0.3,      # bbox[5] - class 0 score at cls_start_idx
    0.7,      # bbox[6] - class 1 score at cls_start_idx + 1
], dtype=np.float32)

# Image size [height, width]
img_size = np.array([480, 640])

# Configuration dictionary with the implicit_background_class key
config = {
    'implicit_background_class': True
}

ARGS = {
    'bbox': bbox,
    'img_size': img_size,
    'model': model,
    'config': config
}
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
bbox_to_ecp_format Ran
ran on a synthesized input
flkraus/bayesian-yolov3/inference_aleatoric.py
code served (permissive licence) · get_code("bdd1738736106015")
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
import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a minimal model with the required attributes
class MinimalModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.cls_start_idx = 9
        self.cls_cnt = 2
        self.obj_idx = 4
    
    def forward(self, x):
        return x

# Create the bbox array with required indices
# bbox needs at least: [0]=y0_norm, [1]=x0_norm, [2]=y1_norm, [3]=x1_norm, 
# [4]=x_var, [5]=y_var, [6]=w_var, [7]=h_var, [8]=total_var,
# [obj_idx]=obj_score, [obj_idx+1]=obj_entropy,
# [cls_start_idx:cls_start_idx+cls_cnt]=cls_scores,
# [cls_start_idx+cls_cnt]=cls_entropy
model = MinimalModel()

# Build bbox array with all required elements
bbox = np.zeros(12, dtype=np.float32)
bbox[0] = 0.1  # y0 normalized
bbox[1] = 0.2  # x0 normalized
bbox[2] = 0.5  # y1 normalized
bbox[3] = 0.7  # x1 normalized
bbox[4] = 0.01  # x_var
bbox[5] = 0.02  # y_var
bbox[6] = 0.03  # w_var
bbox[7] = 0.04  # h_var
bbox[8] = 0.1   # total_var
bbox[model.obj_idx] = 0.8  # obj_score (index 4)
bbox[model.obj_idx + 1] = 0.15  # obj_entropy (index 5)
bbox[model.cls_start_idx:model.cls_start_idx + model.cls_cnt] = [0.9, 0.1]  # cls_scores (indices 9-10)
bbox[model.cls_start_idx + model.cls_cnt] = 0.5  # cls_entropy (index 11)

img_size = (480, 640, 3)

config = {
    'implicit_background_class': True
}

ARGS = {
    'bbox': bbox,
    'img_size': img_size,
    'model': model,
    'config': config
}
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
bbox_to_ecp_format Ran
ran on a synthesized input
flkraus/bayesian-yolov3/inference_epistemic.py
code served (permissive licence) · get_code("8a5715edef2f0423")
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
import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a minimal model class that provides the attributes used in the function
class MinimalModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.cls_start_idx = 14
        self.cls_cnt = 2
        self.obj_idx = 3
    
    def forward(self, x):
        return x

# Create a minimal config dict
config = {
    'implicit_background_class': True
}

# Create a bbox with all required indices
# Indices 0-3: y0, x0, y1, x1 (normalized coordinates)
# Indices 4-7: x_var_epi, y_var_epi, w_var_epi, h_var_epi
# Indices 8-11: x_var_ale, y_var_ale, w_var_ale, h_var_ale
# Indices 12-13: total_var_epi, total_var_ale
# Index 3 (obj_idx): objectness score
# Indices 4-5 (obj_idx+1, obj_idx+2): obj_mutual_info, obj_entropy
# Indices 14-15 (cls_start_idx): cls_scores for 2 classes
# Indices 16-17: ped_score, rider_score (at indices 17-18 in output, but accessed as bbox[17], bbox[18])
# Indices 16-19 (cls_start_idx + cls_cnt to cls_start_idx + cls_cnt + 3): cls_mutual_info, cls_entropy, layer_id, prior_id
bbox = np.array([
    0.1,    # 0: y0 (normalized)
    0.2,    # 1: x0 (normalized)
    0.5,    # 2: y1 (normalized)
    0.7,    # 3: x1 (normalized) -- also obj_idx
    0.01,   # 4: x_var_epi
    0.02,   # 5: y_var_epi
    0.03,   # 6: w_var_epi
    0.04,   # 7: h_var_epi
    0.05,   # 8: x_var_ale
    0.06,   # 9: y_var_ale
    0.07,   # 10: w_var_ale
    0.08,   # 11: h_var_ale
    0.15,   # 12: total_var_epi
    0.25,   # 13: total_var_ale
    0.8,    # 14: cls_scores[0] (pedestrian)
    0.2,    # 15: cls_scores[1] (rider)
    0.9,    # 16: ped_score
    0.85,   # 17: rider_score (accessed as bbox[18], so we need one more entry)
    0.12,   # 18: cls_mutual_info (cls_start_idx + cls_cnt = 14 + 2 = 16... wait)
], dtype=np.float32)

# Recalculate: cls_start_idx=14, cls_cnt=2
# cls_scores are at indices 14-15
# obj_idx = 3
# obj_mutual_info at obj_idx+1 = 4
# obj_entropy at obj_idx+2 = 5
# ped_score at bbox[17]
# rider_score at bbox[18]
# cls_mutual_info at cls_start_idx + cls_cnt = 16
# cls_entropy at cls_start_idx + cls_cnt + 1 = 17
# layer_id at cls_start_idx + cls_cnt + 2 = 18
# prior_id at cls_start_idx + cls_cnt + 3 = 19

# Rebuild bbox with correct structure
bbox = np.array([
    0.1,    # 0: y0
    0.2,    # 1: x0
    0.5,    # 2: y1
    0.7,    # 3: x1 / obj_idx
    0.01,   # 4: x_var_epi / obj_mutual_info
    0.02,   # 5: y_var_epi / obj_entropy
    0.03,   # 6: w_var_epi
    0.04,   # 7: h_var_epi
    0.05,   # 8: x_var_ale
    0.06,   # 9: y_var_ale
    0.07,   # 10: w_var_ale
    0.08,   # 11: h_var_ale
    0.15,   # 12: total_var_epi
    0.25,   # 13: total_var_ale
    0.8,    # 14: cls_scores[0]
    0.2,    # 15: cls_scores[1]
    0.12,   # 16: cls_mutual_info
    0.85,   # 17: ped_score / cls_entropy
    0.9,    # 18: rider_score / layer_id
    0.05,   # 19: prior_id
], dtype=np.float32)

img_size = (480, 640, 3)
model = MinimalModel()

ARGS = {
    'bbox': bbox,
    'img_size': img_size,
    'model': model,
    'config': config
}
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
bytes_feature Ran
ran; our contract draft was wrong, not the code
flkraus/bayesian-yolov3/create_tf_records_citypersons.py
code served (permissive licence) · get_code("d682a9593710857e")
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 Feature. 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 tensorflow as tf

# bytes_feature expects a single parameter 'value' which should be bytes
# The function wraps it in tf.train.BytesList and then tf.train.Feature

ARGS = {
    "value": b"example_bytes_data"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Feature — a Feature output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Feature 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
concat_bbox Ran
ran on a synthesized input
flkraus/bayesian-yolov3/inference_standard_yolov3.py
code served (permissive licence) · get_code("6ba1ae5ae4da3d61")
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 EagerTensor, ndim 3, dtype family float, shape (2, 36, 5) 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 tensorflow as tf
import numpy as np

# Create a mock net_out structure that the function iterates over
# The function expects: net_out -> det_layer -> prior with shape [batches, lw, lh, det_size]

# Create sample prior tensors with distinct dimensions
# batches=2, lw=3, lh=4, det_size=5
prior1 = tf.constant(np.random.randn(2, 3, 4, 5), dtype=tf.float32)
prior2 = tf.constant(np.random.randn(2, 3, 4, 5), dtype=tf.float32)

# Create det_layer as list of priors
det_layer1 = [prior1, prior2]
det_layer2 = [prior1]

# Create net_out as list of detection layers
net_out = [det_layer1, det_layer2]

ARGS = {
    "net_out": net_out
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 3, dtype family float, shape (2, 36, 5) — shape (2, 36, 5) 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
int64_feature Ran
ran; our contract draft was wrong, not the code
flkraus/bayesian-yolov3/create_tf_records_citypersons.py
code served (permissive licence) · get_code("328d17201d9a0e68")
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 Feature. 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 tensorflow as tf

def int64_feature(value):
    return tf.train.Feature(int64_list=tf.train.Int64List(value=[value]))

ARGS = {
    "value": 42
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Feature — a Feature output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Feature 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
int64_list_feature Ran
ran on a synthesized input
flkraus/bayesian-yolov3/create_tf_records_citypersons.py
code served (permissive licence) · get_code("f6d735872f4495ed")
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 Feature. 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 tensorflow as tf

# The function expects a 'value' parameter that will be passed to
# tf.train.Int64List(value=value). This should be an iterable of integers.

ARGS = {
    "value": [1, 2, 3, 4, 5]
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Feature — a Feature output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Feature 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
flkraus/bayesian-yolov3/inference_standard_yolov3.py
code served (permissive licence) · get_code("7d36f19ae2c4eeb2")
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 EagerTensor, ndim 3, dtype family float, shape (2, 2, 5) 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 tensorflow as tf

class MockModel:
    """Minimal model class that provides the obj_idx attribute used in nms_op."""
    def __init__(self, obj_idx=4):
        self.obj_idx = obj_idx

# Create a mock model with obj_idx pointing to the confidence score column
model = MockModel(obj_idx=4)

# Create sample boxes tensor
# Shape: (batch_size=2, num_boxes=3, features=5)
# Features: [x1, y1, x2, y2, confidence_score]
# We need boxes with at least 5 columns (4 for coordinates + 1 for confidence at index 4)
all_boxes = tf.constant([
    [
        [10.0, 20.0, 30.0, 40.0, 0.9],
        [15.0, 25.0, 35.0, 45.0, 0.8],
        [12.0, 22.0, 32.0, 42.0, 0.7],
    ],
    [
        [50.0, 60.0, 70.0, 80.0, 0.95],
        [55.0, 65.0, 75.0, 85.0, 0.85],
        [52.0, 62.0, 72.0, 82.0, 0.75],
    ]
], dtype=tf.float32)

ARGS = {
    "all_boxes": all_boxes,
    "model": model,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 3, dtype family float, shape (2, 2, 5) — shape (2, 2, 5) 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
nms Ran
ran on a synthesized input
flkraus/bayesian-yolov3/inference_epistemic.py
code served (permissive licence) · get_code("2211cad211bfdc92")
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 EagerTensor, ndim 2, dtype family float, shape (2, 5) 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 tensorflow as tf
import numpy as np

# Create a small stand-in model class
class MockModel:
    def __init__(self):
        self.obj_idx = 4  # Index for objectness scores in boxes tensor

model = MockModel()

# Create boxes tensor: shape [2, 5]
# First 4 columns are box coordinates, 5th column is objectness score
boxes_data = np.array([
    [10.0, 20.0, 30.0, 40.0, 0.9],
    [15.0, 25.0, 35.0, 45.0, 0.8]
], dtype=np.float32)
boxes = tf.constant(boxes_data)

ARGS = {
    "boxes": boxes,
    "model": model
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 2, dtype family float, shape (2, 5) — shape (2, 5) 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
split_detection Ran
ran on a synthesized input
flkraus/bayesian-yolov3/lib_yolo/layers.py
code served (permissive licence) · get_code("05a04abe14c59884")
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 tensorflow as tf

# The function splits detection outputs into localization, objectness, and class predictions
# It expects:
# - inputs: a tensor with shape (batch, ..., boxes_per_cell * (4 + 1 + cls_cnt))
# - boxes_per_cell: number of boxes per cell (integer)
# - cls_cnt: number of classes (integer)

# Create a small batch of detection data
# Shape: (batch_size=2, spatial_dim=3, total_features=boxes_per_cell * (4 + 1 + cls_cnt))
batch_size = 2
spatial_dim = 3
boxes_per_cell = 2
cls_cnt = 5

# Total features per location = boxes_per_cell * (4 loc + 1 obj + cls_cnt)
total_features = boxes_per_cell * (4 + 1 + cls_cnt)

# Create input tensor with shape (batch_size, spatial_dim, total_features)
inputs = tf.random.normal([batch_size, spatial_dim, total_features])

ARGS = {
    'inputs': inputs,
    'boxes_per_cell': boxes_per_cell,
    'cls_cnt': cls_cnt
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — 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
split_detection_aleatoric Ran
ran on a synthesized input
flkraus/bayesian-yolov3/lib_yolo/layers.py
code served (permissive licence) · get_code("5bde5713476f5229")
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 tensorflow as tf

# Set random seed for reproducibility
tf.random.set_seed(0)

# Define the arguments for split_detection_aleatoric
# The function expects:
# - inputs: a tensor that will be split by boxes_per_cell
# - boxes_per_cell: number of boxes per cell (determines first split)
# - cls_cnt: number of classes

# Each box detection has: 4 (loc) + 4 (log_loc_var) + 1 (obj) + 1 (log_obj_stddev) + cls_cnt (cls) + cls_cnt (log_cls_stddev)
# Total per box = 4 + 4 + 1 + 1 + cls_cnt + cls_cnt = 10 + 2*cls_cnt

boxes_per_cell = 2
cls_cnt = 3

# Per-box feature size
per_box_size = 4 + 4 + 1 + 1 + cls_cnt + cls_cnt  # = 16

# Total last dimension = boxes_per_cell * per_box_size
total_last_dim = boxes_per_cell * per_box_size  # = 32

# Create a small batch of inputs
# Shape: [batch_size, height, width, total_last_dim] or just [batch_size, total_last_dim]
# Using a simple 2D input for clarity: [batch_size, feature_dim]
batch_size = 2
inputs = tf.random.normal([batch_size, total_last_dim])

ARGS = {
    'inputs': inputs,
    'boxes_per_cell': boxes_per_cell,
    'cls_cnt': cls_cnt
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — 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
aleatoric_obj_loss Not yet run
unverified (harvested, no recorded run)
flkraus/bayesian-yolov3/lib_yolo/layers.py
code served (permissive licence) · get_code("8b895c78336b61c6")
How it ran

no invocation recorded.

load_darknet_weights Not yet run
unverified (harvested, no recorded run)
flkraus/bayesian-yolov3/lib_yolo/darknet.py
code served (permissive licence) · get_code("281d23c0cd732c87")
How it ran

no invocation recorded.

Also present in other papers' code

Of 13 samples lifted from this paper's linked repositories, 13 carry a function identity and 0 do not. Of the 13 distinct functions with an identity, 3 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 0 the same month, 3 later) and 10 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
bytes_feature 8 other papers
7 dated earlier, 1 later
1409.1556 (2014-09)
Very Deep Convolutional Networks for Large-Scale Image Recognition
1607.08022 (2016-07)
Instance Normalization: The Missing Ingredient for Fast Stylization
1609.05158 (2016-09)
Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network
and 5 more not named here
int64_feature 7 other papers
6 dated earlier, 1 later
1607.08022 (2016-07)
Instance Normalization: The Missing Ingredient for Fast Stylization
1611.07810 (2016-11)
A dataset and exploration of models for understanding video data through fill-in-the-blank question-answering
1802.05591 (2018-02)
Towards End-to-End Lane Detection: an Instance Segmentation Approach
and 4 more not named here
int64_list_feature 5 other papers
4 dated earlier, 1 later
1607.08022 (2016-07)
Instance Normalization: The Missing Ingredient for Fast Stylization
1803.10562 (2018-03)
ELEGANT: Exchanging Latent Encodings with GAN for Transferring Multiple Face Attributes
1812.00442 (2018-12)
Deep Cosine Metric Learning for Person Re-Identification
and 2 more not named here

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Environment perception is the task for intelligent vehicles on which all subsequent steps rely. A key part of perception is to safely detect other road users such as vehicles, pedestrians, and cyclists. With modern deep learning techniques huge progress was made over the last years in this field. However such deep learning based object detection models cannot predict how certain they are in their predictions, potentially hampering the performance of later steps such as tracking or sensor fusion. We present a viable approaches to estimate uncertainty in an one-stage object detector, while improving the detection performance of the baseline approach. The proposed model is evaluated on a large scale automotive pedestrian dataset. Experimental results show that the uncertainty outputted by our system is coupled with detection accuracy and the occlusion level of pedestrians.

For agents

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

get_harvested_code_for_paper("1905.10296")
get_code_for_paper("1905.10296")
have("1905.10296")

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

curl https://syntology.ai/api/ran/1905.10296.json

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

[![Syntology run record](https://syntology.ai/api/ran/1905.10296.svg)](https://syntology.ai/paper/1905.10296)

Connect an agent — have() is free.