SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2104.09224 · CVPR · 2021

Multi-Modal Fusion Transformer for End-to-End Autonomous Driving

Andreas Geiger, Kashyap Chitta, Aditya Prakash

arXiv · PDF · Open in the Atlas

Code that ran

9 of 15 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (7 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 1 violated it and 7 have no contract verdict; 1 failure of our instrument; 6 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/2104.09224.json · badge: /api/ran/2104.09224.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
Kin-Zhang/mmfn community
Papers with Code archive
6 ran · 3 no recorded run
autonomousvision/transfuser marked official in Papers with Code
Papers with Code archive
3 ran · 3 no recorded run
FunctionStatusWhere it lives
Block Ran
ran on a synthesized input · constructed an object; no result was computed
Kin-Zhang/mmfn/team_code/benchmarks/transfuser/model.py
code served (permissive licence) · get_code("5075118d0c207c37")
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 an instance of its own class, Block: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "n_embd": 8,
    "n_head": 2,
    "block_exp": 3,
    "attn_pdrop": 0.1,
    "resid_pdrop": 0.1,
}
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 Block — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Block 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
Block Ran
ran on a synthesized input · constructed an object; no result was computed
autonomousvision/transfuser/team_code_transfuser/transfuser.py
code served (permissive licence) · get_code("d93ea624e09cfe3d")
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 an instance of its own class, Block: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn
import math

torch.manual_seed(0)

ARGS = {
    "n_embd": 8,
    "n_head": 2,
    "block_exp": 4,
    "attn_pdrop": 0.1,
    "resid_pdrop": 0.1,
}
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 Block — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Block 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
GPT Ran
ran on a synthesized input · constructed an object; no result was computed
Kin-Zhang/mmfn/team_code/benchmarks/transfuser/model.py
code served (permissive licence) · get_code("594894f4b9eb35a1")
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 an instance of its own class, GPT: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

# Create a minimal config class with required attributes
class Config:
    def __init__(self):
        self.n_views = 2

config = Config()

# Parameters for GPT
n_embd = 8
n_head = 2
block_exp = 2
n_layer = 1
vert_anchors = 2
horz_anchors = 2
seq_len = 2
embd_pdrop = 0.1
attn_pdrop = 0.1
resid_pdrop = 0.1

# Construct the config object as required
gpt_config = Config()

ARGS = {
    "n_embd": n_embd,
    "n_head": n_head,
    "block_exp": block_exp,
    "n_layer": n_layer,
    "vert_anchors": vert_anchors,
    "horz_anchors": horz_anchors,
    "seq_len": seq_len,
    "embd_pdrop": embd_pdrop,
    "attn_pdrop": attn_pdrop,
    "resid_pdrop": resid_pdrop,
    "config": gpt_config,
}
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 GPT — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a GPT 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
GPT Ran
ran on a synthesized input · constructed an object; no result was computed
autonomousvision/transfuser/team_code_transfuser/transfuser.py
code served (permissive licence) · get_code("7a6582011815aa73")
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 an instance of its own class, GPT: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn
import math

torch.manual_seed(0)

# Stand-in config class with the attributes used in GPT._init_weights
class Config:
    def __init__(self):
        self.gpt_linear_layer_init_mean = 0.0
        self.gpt_linear_layer_init_std = 0.02
        self.gpt_layer_norm_init_weight = 1.0

config = Config()

# Parameters for GPT
n_embd = 8
n_head = 2
block_exp = 2
n_layer = 1
img_vert_anchors = 3
img_horz_anchors = 4
lidar_vert_anchors = 2
lidar_horz_anchors = 3
seq_len = 1
embd_pdrop = 0.1
attn_pdrop = 0.1
resid_pdrop = 0.1
use_velocity = True

ARGS = {
    "n_embd": n_embd,
    "n_head": n_head,
    "block_exp": block_exp,
    "n_layer": n_layer,
    "img_vert_anchors": img_vert_anchors,
    "img_horz_anchors": img_horz_anchors,
    "lidar_vert_anchors": lidar_vert_anchors,
    "lidar_horz_anchors": lidar_horz_anchors,
    "seq_len": seq_len,
    "embd_pdrop": embd_pdrop,
    "attn_pdrop": attn_pdrop,
    "resid_pdrop": resid_pdrop,
    "config": config,
    "use_velocity": use_velocity,
}
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 GPT — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a GPT 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
LidarEncoder Ran
ran on a synthesized input · constructed an object; no result was computed
Kin-Zhang/mmfn/team_code/benchmarks/transfuser/model.py
code served (permissive licence) · get_code("152cd035838391cd")
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 an instance of its own class, LidarEncoder: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn
from torchvision import models

torch.manual_seed(0)

# The LidarEncoder.forward() method expects:
# - inputs: an iterable of tensors (lidar_data)
# - each lidar_data is passed through self._model which is a ResNet18
# - ResNet18 expects 4D input: (batch, channels, height, width)
# - The modified conv1 expects in_channels input channels
# - The output features are accumulated via addition

# Create sample lidar data - list of tensors
# Each tensor should have shape (batch, in_channels, height, width)
# Using batch=2, in_channels=2, height=4, width=3 for distinct dimensions
in_channels = 2
lidar_data_1 = torch.randn(2, in_channels, 4, 3)
lidar_data_2 = torch.randn(2, in_channels, 4, 3)
lidar_inputs = [lidar_data_1, lidar_data_2]

ARGS = {
    "num_classes": 512,
    "in_channels": 2,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type LidarEncoder — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a LidarEncoder 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
PIDController Ran
ran on a synthesized input · constructed an object; no result was computed
Kin-Zhang/mmfn/team_code/benchmarks/transfuser/model.py
code served (permissive licence) · get_code("bd73b9145f2b9944")
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 an instance of its own class, PIDController: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
from collections import deque

ARGS = {
    'K_P': 1.5,
    'K_I': 0.1,
    'K_D': 0.05,
    'n': 20
}
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 PIDController — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PIDController 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
SelfAttention Ran
ran on a synthesized input · constructed an object; no result was computed
Kin-Zhang/mmfn/team_code/benchmarks/transfuser/model.py
code served (permissive licence) · get_code("6141dbd1107c57fd")
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 an instance of its own class, SelfAttention: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn
import math

torch.manual_seed(0)

# Parameters for SelfAttention
n_embd = 8  # embedding dimension, must be divisible by n_head
n_head = 2  # number of attention heads
attn_pdrop = 0.1  # attention dropout probability
resid_pdrop = 0.1  # residual dropout probability

ARGS = {
    "n_embd": n_embd,
    "n_head": n_head,
    "attn_pdrop": attn_pdrop,
    "resid_pdrop": resid_pdrop,
}
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 SelfAttention — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a SelfAttention 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
normalize_imagenet Ran
ran, violated the contract we drafted
Kin-Zhang/mmfn/team_code/benchmarks/transfuser/model.py
code served (permissive licence) · get_code("af36a314f4f4a4be")
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 4, dtype family float, shape (2, 3, 4, 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)

# The function expects a tensor x with shape (batch_size, channels, height, width)
# where channels=3 (RGB). It accesses x[:, 0], x[:, 1], x[:, 2] for the three channels.
# The tensor should contain normalized values in a reasonable range (e.g., [0, 1] or similar).

x = torch.randn(2, 3, 4, 4)  # batch=2, channels=3, height=4, width=4

ARGS = {
    "x": x
}
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 4, dtype family float, shape (2, 3, 4, 4) — shape (2, 3, 4, 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
normalize_imagenet Ran
ran; our contract draft was wrong, not the code
autonomousvision/transfuser/team_code_transfuser/transfuser.py
code served (permissive licence) · get_code("1529467107cafc96")
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 4, dtype family float, shape (2, 3, 4, 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 batch of images with shape (batch_size, channels, height, width)
# The function indexes x[:, 0], x[:, 1], x[:, 2], so we need 3 channels
# Values should be in range [0, 255] since the function divides by 255.0
x = torch.randint(0, 256, (2, 3, 4, 4), dtype=torch.float32)

ARGS = {
    "x": x
}
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 4, dtype family float, shape (2, 3, 4, 4) — shape (2, 3, 4, 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
Encoder Not yet run
unverified (harvested, no recorded run)
Kin-Zhang/mmfn/team_code/benchmarks/transfuser/model.py
code served (permissive licence) · get_code("277c052f581f3276")
How it ran

no invocation recorded.

ImageCNN Not yet run
unverified (harvested, no recorded run)
Kin-Zhang/mmfn/team_code/benchmarks/transfuser/model.py
code served (permissive licence) · get_code("1a4dc7d1e43090ab")
How it ran

no invocation recorded.

ImageCNN Not yet run
unverified (harvested, no recorded run)
autonomousvision/transfuser/team_code_transfuser/transfuser.py
code served (permissive licence) · get_code("e1a45fd31eaf668d")
How it ran

no invocation recorded.

LidarEncoder Not yet run
unverified (harvested, no recorded run)
autonomousvision/transfuser/team_code_transfuser/transfuser.py
code served (permissive licence) · get_code("d2d93b6e784851f4")
How it ran

no invocation recorded.

TransFuser Not yet run
unverified (harvested, no recorded run)
Kin-Zhang/mmfn/team_code/benchmarks/transfuser/model.py
code served (permissive licence) · get_code("3d3c28ac7ec98753")
How it ran

no invocation recorded.

TransfuserBackbone Not yet run
unverified (harvested, no recorded run)
autonomousvision/transfuser/team_code_transfuser/transfuser.py
code served (permissive licence) · get_code("8d4dbe093612822d")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 distinct functions with an identity, 5 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 5 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
normalize_imagenet
brick of this paper's assembly: helper
7 other papers
2 dated earlier, 5 later · 1 only as an assembly brick
1911.05063 (2019-11)
Kaolin: A PyTorch Library for Accelerating 3D Deep Learning Research
2007.10973 (2020-07)
Neural Mesh Flow: 3D Manifold Mesh Generation via Diffeomorphic Flows
2106.02634 (2021-06)
Light Field Networks: Neural Scene Representations with Single-Evaluation Rendering
and 4 more not named here
Block
brick of this paper's assembly: model
2 other papers
2 later
2306.07957 (2023-06)
Hidden Biases of End-to-End Driving Models
2504.01941 (2025-04)
End-to-End Driving with Online Trajectory Evaluation via BEV World Model
Block
brick of this paper's assembly: model
1 other paper
1 later · 1 only as an assembly brick
2109.04456 (2021-09)
NEAT: Neural Attention Fields for End-to-End Autonomous Driving
SelfAttention
brick of this paper's assembly: model
1 other paper
1 later · 1 only as an assembly brick
2109.04456 (2021-09)
NEAT: Neural Attention Fields for End-to-End Autonomous Driving
normalize_imagenet
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2306.07957 (2023-06)
Hidden Biases of End-to-End Driving Models

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

How should representations from complementary sensors be integrated for autonomous driving? Geometrybased sensor fusion has shown great promise for perception tasks such as object detection and motion forecasting. However, for the actual driving task, the global context of the 3D scene is key, e.g. a change in traffic light state can affect the behavior of a vehicle geometrically distant from that traffic light. Geometry alone may therefore be insufficient for effectively fusing representations in end-to-end driving models. In this work, we demonstrate that imitation learning policies based on existing sensor fusion methods under-perform in the presence of a high density of dynamic agents and complex scenarios, which require global contextual reasoning, such as handling traffic oncoming from multiple directions at uncontrolled intersections. Therefore, we propose TransFuser, a novel Multi-Modal Fusion Transformer, to integrate image and LiDAR representations using attention. We experimentally validate the efficacy of our approach in urban settings involving complex scenarios using the CARLA urban driving simulator. Our approach achieves state-of-the-art driving performance while reducing collisions by 76% compared to geometry-based fusion.

For agents

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

get_harvested_code_for_paper("2104.09224")
get_code_for_paper("2104.09224")
have("2104.09224")

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

curl https://syntology.ai/api/ran/2104.09224.json

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

[![Syntology run record](https://syntology.ai/api/ran/2104.09224.svg)](https://syntology.ai/paper/2104.09224)

Connect an agent — have() is free.