SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · aaai_28449 · AAAI · 2024

Regulating Intermediate 3D Features for Vision-Centric Autonomous Driving

Deng Cai, Wenxiao Wang, Liang Peng, Qi Zhou, Haoran Cheng, Junkai Xu, Linxuan Xia, Dan Deng, Wei Qian

Open in the Atlas

Code that ran

7 of 9 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; 1 honoured the contract we drafted, 1 violated it and 1 has no contract verdict; 4 failures of our instrument; 2 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/aaai_28449.json · badge: /api/ran/aaai_28449.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
cskkxjk/Vampire canonical (classifier label)
classifier basis: census
7 ran · 2 no recorded run
FunctionStatusWhere it lives
all_gather_object Ran
ran; our contract draft was wrong, not the code
cskkxjk/Vampire/src/utils/torch_dist.py
pointer only (licence: MIT); the stored copy get_code serves, from datongjie/bevspread, is pointer only (licence: NONE) · get_code("07a0962477fe7ed9")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from datongjie/bevspread, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_rot Ran
ran, honoured the contract we drafted
cskkxjk/Vampire/src/datasets/nusc_det_seg_dataset.py
code served (permissive licence) · get_code("2dd096050170984b")
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, 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
import torch

torch.manual_seed(0)

# The function get_rot takes a single parameter h (a scalar angle)
# and returns a 2x2 rotation matrix as a torch.Tensor
# We need to provide a valid angle value

h = np.pi / 4  # 45 degrees, a valid angle for rotation

ARGS = {
    "h": h
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 2) — shape (2, 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
img_transform Ran
ran; our contract draft was wrong, not the code
cskkxjk/Vampire/src/datasets/nusc_det_seg_dataset.py
code served (permissive licence) · get_code("42061d7d8a0c3f80")
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
from PIL import Image

torch.manual_seed(0)

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

# Define all parameters
resize = 0.5
resize_dims = (50, 50)
crop = (10, 10, 40, 40)
flip = True
rotate = 15.0

ARGS = {
    'img': img,
    'resize': resize,
    'resize_dims': resize_dims,
    'crop': crop,
    'flip': flip,
    'rotate': rotate
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, 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
iou Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
cskkxjk/Vampire/src/utils/lovasz_losses.py
code served (permissive licence) · get_code("d94639b28b2134ff")
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 (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
import torch

torch.manual_seed(0)

# Construct valid arguments for the iou function
# preds: prediction array/tensor with class indices
# labels: ground truth array/tensor with class indices
# C: number of classes
# EMPTY: value to return for empty unions (default 1.0)
# ignore: class index to ignore (default None)
# per_image: whether to compute per-image IoU (default False)

# Create small prediction and label arrays
# Using shape (2, 4) - batch of 2, spatial dimension of 4
preds = np.array([[0, 1, 2, 3],
                   [1, 2, 0, 1]], dtype=np.int64)

labels = np.array([[0, 1, 1, 3],
                    [1, 2, 2, 1]], dtype=np.int64)

# Number of classes
C = 4

# EMPTY value for classes with no union
EMPTY = 1.0

# Class to ignore during computation
ignore = None

# Whether to compute per-image IoU
per_image = False

ARGS = {
    "preds": preds,
    "labels": labels,
    "C": C,
    "EMPTY": EMPTY,
    "ignore": ignore,
    "per_image": per_image
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (4,) — shape (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
iou_binary Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
cskkxjk/Vampire/src/utils/lovasz_losses.py
code served (permissive licence) · get_code("1211530cd527bd0b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

def isnan(x):
    return x != x

def mean(l, ignore_nan=False, empty=0):
    """
    nanmean compatible with generators.
    """
    l = iter(l)
    if ignore_nan:
        from itertools import filterfalse
        l = filterfalse(isnan, l)
    try:
        n = 1
        acc = next(l)
    except StopIteration:
        if empty == 'raise':
            raise ValueError('Empty mean')
        return empty
    for n, v in enumerate(l, 2):
        acc += v
    if n == 1:
        return acc
    return acc / n

def iou_binary(preds, labels, EMPTY=1.0, ignore=None, per_image=True):
    """
    IoU for foreground class
    binary: 1 foreground, 0 background
    """
    if not per_image:
        preds, labels = ((preds,), (labels,))
    ious = []
    for pred, label in zip(preds, labels):
        intersection = ((label == 1) & (pred == 1)).sum()
        union = ((label == 1) | (pred == 1) & (label != ignore)).sum()
        if not union:
            iou = EMPTY
        else:
            iou = float(intersection) / float(union)
        ious.append(iou)
    iou = mean(ious)
    return 100 * iou

# Setup code
np.random.seed(0)

# Create small prediction and label arrays for binary segmentation
# Shape: (height=4, width=3) for two images
preds_img1 = np.array([[0, 1, 1],
                       [1, 1, 0],
                       [0, 1, 1],
                       [1, 0, 0]], dtype=np.int32)

labels_img1 = np.array([[0, 1, 0],
                        [1, 1, 1],
                        [0, 1, 0],
                        [1, 0, 1]], dtype=np.int32)

preds_img2 = np.array([[1, 1, 0],
                       [0, 1, 1],
                       [1, 1, 0],
                       [0, 0, 1]], dtype=np.int32)

labels_img2 = np.array([[1, 0, 0],
                        [0, 1, 1],
                        [1, 0, 0],
                        [0, 1, 1]], dtype=np.int32)

# Arguments for iou_binary function
ARGS = {
    'preds': np.array([preds_img1, preds_img2]),
    'labels': np.array([labels_img1, labels_img2]),
    'EMPTY': 1.0,
    'ignore': 0,
    'per_image': 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 float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
lovasz_grad Ran
ran, violated the contract we drafted
cskkxjk/Vampire/src/utils/lovasz_losses.py
code served (permissive licence) · get_code("e85b5d21dc5b8382")
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 1, dtype family float, shape (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)

# gt_sorted should be a 1D tensor of binary/boolean values
# representing ground truth sorted by some error metric
# The function expects a tensor with sum(), float(), cumsum() methods
gt_sorted = torch.tensor([0, 0, 1, 1], dtype=torch.long)

ARGS = {
    "gt_sorted": gt_sorted
}
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 1, dtype family float, shape (4,) — shape (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
run_time Ran
ran on a synthesized input
cskkxjk/Vampire/src/utils/bricks.py
code served (permissive licence) · get_code("65fc1ebf7e78cfb6")
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 function. 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 time
from collections import defaultdict
import torch

torch.manual_seed(0)

time_maps = defaultdict(lambda: 0.0)
count_maps = defaultdict(lambda: 0.0)

def run_time(name):
    def middle(fn):
        def wrapper(*args, **kwargs):
            torch.cuda.synchronize()
            start = time.time()
            res = fn(*args, **kwargs)
            torch.cuda.synchronize()
            time_maps['%s : %s' % (name, fn.__name__)] += time.time() - start
            count_maps['%s : %s' % (name, fn.__name__)] += 1
            print('%s : %s takes up %f ' % (name, fn.__name__, time_maps['%s : %s' % (name, fn.__name__)] / count_maps['%s : %s' % (name, fn.__name__)]))
            return res
        return wrapper
    return middle

# Test the decorator
@run_time("test_operation")
def sample_function(x, y):
    return x + y

# Call it to verify it works
result = sample_function(torch.randn(2, 4), torch.randn(2, 4))
result = sample_function(torch.randn(2, 4), torch.randn(2, 4))

ARGS = {
    "name": "forward_pass"
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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
bev_transform Not yet run
unverified (harvested, no recorded run)
cskkxjk/Vampire/src/datasets/nusc_det_seg_dataset.py
code served (permissive licence) · get_code("f7f40a523ab29daf")
How it ran

no invocation recorded.

is_parallel Not yet run
unverified (harvested, no recorded run)
cskkxjk/Vampire/src/callbacks/ema.py
code served (permissive licence) · get_code("bbbb5d6a28074b9d")
How it ran

no invocation recorded.

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 distinct functions with an identity, 7 are also present in other papers whose code we hold (7 with at least one paper dated earlier, 0 the same month, 3 later) and 2 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
lovasz_grad 40 other papers
28 dated earlier, 5 later, 7 same year, order not determinable · 1 only as an assembly brick
1505.04597 (2015-05)
U-Net: Convolutional Networks for Biomedical Image Segmentation
1807.10165 (2018-07)
UNet++: A Nested U-Net Architecture for Medical Image Segmentation
1811.11721 (2018-11)
CCNet: Criss-Cross Attention for Semantic Segmentation
and 37 more not named here
iou 24 other papers
16 dated earlier, 2 later, 6 same year, order not determinable
1807.10165 (2018-07)
UNet++: A Nested U-Net Architecture for Medical Image Segmentation
1811.11721 (2018-11)
CCNet: Criss-Cross Attention for Semantic Segmentation
2003.10778 (2020-03)
PanNuke Dataset Extension, Insights and Baselines
and 21 more not named here
iou_binary 24 other papers
16 dated earlier, 2 later, 6 same year, order not determinable
1807.10165 (2018-07)
UNet++: A Nested U-Net Architecture for Medical Image Segmentation
1811.11721 (2018-11)
CCNet: Criss-Cross Attention for Semantic Segmentation
2003.10778 (2020-03)
PanNuke Dataset Extension, Insights and Baselines
and 21 more not named here
is_parallel 8 other papers
8 dated earlier
1803.11365 (2018-03)
Cross-Domain Weakly-Supervised Object Detection through Progressive Domain Adaptation
2207.02541 (2022-07)
Dense Teacher: Dense Pseudo-Labels for Semi-supervised Object Detection
2209.10248 (2022-09)
BEVStereo: Enhancing Depth Estimation in Multi-view 3D Object Detection with Dynamic Temporal Stereo
and 5 more not named here
get_rot 5 other papers
5 dated earlier
2209.10248 (2022-09)
BEVStereo: Enhancing Depth Estimation in Multi-view 3D Object Detection with Dynamic Temporal Stereo
2211.09445 (2022-11)
aiMotive Dataset: A Multimodal Dataset for Robust Autonomous Driving with Long-Range Perception
2211.12501 (2022-11)
AeDet: Azimuth-invariant Multi-view 3D Object Detection
and 2 more not named here
img_transform 5 other papers
5 dated earlier
2209.10248 (2022-09)
BEVStereo: Enhancing Depth Estimation in Multi-view 3D Object Detection with Dynamic Temporal Stereo
2211.09445 (2022-11)
aiMotive Dataset: A Multimodal Dataset for Robust Autonomous Driving with Long-Range Perception
2211.12501 (2022-11)
AeDet: Azimuth-invariant Multi-view 3D Object Detection
and 2 more not named here
all_gather_object 2 other papers
1 dated earlier, 1 same year, order not determinable
2303.08498 (2023-03)
BEVHeight: A Robust Framework for Vision-based Roadside 3D Object Detection
2406.08785 (2024-06)
BEVSpread: Spread Voxel Pooling for Bird's-Eye-View Representation in Vision-based Roadside 3D Object Detection

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

Multi-camera perception tasks have gained significant attention in the field of autonomous driving. However, existing frameworks based on Lift-Splat-Shoot (LSS) in the multicamera setting cannot produce suitable dense 3D features due to the projection nature and uncontrollable densification process. To resolve this problem, we propose to regulate intermediate dense 3D features with the help of volume rendering. Specifically, we employ volume rendering to process the dense 3D features to obtain corresponding 2D features (e.g., depth maps, semantic maps), which are supervised by associated labels in the training. This manner regulates the generation of dense 3D features on the feature level, providing appropriate dense and unified features for multiple perception tasks. Therefore, our approach is termed Vampire, stands for "Volume rendering As Multi-camera Perception Intermediate feature REgulator". Experimental results on the Occ3D and nuScenes datasets demonstrate that Vampire facilitates fine-grained and appropriate extraction of dense 3D features, and is competitive with existing SOTA methods across diverse downstream perception tasks like 3D occupancy prediction, LiDAR segmentation and 3D objection detection, while utilizing moderate GPU resources. We provide a video demonstration in the supplementary materials and Codes are available at github.com/cskkxjk/Vampire.

For agents

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

get_harvested_code_for_paper("aaai_28449")
get_code_for_paper("aaai_28449")
have("aaai_28449")

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

curl https://syntology.ai/api/ran/aaai_28449.json

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

[![Syntology run record](https://syntology.ai/api/ran/aaai_28449.svg)](https://syntology.ai/paper/aaai_28449)

Connect an agent — have() is free.