SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2403.09630 · CVPR · 2024

GenAD: Generalized Predictive Model for Autonomous Driving

arXiv · PDF · Open in the Atlas

Code that ran

8 of 8 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 2 have no contract verdict; 4 failures of our instrument; 0 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/2403.09630.json · badge: /api/ran/2403.09630.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
opendrivelab/driveagi canonical (classifier label)
classifier basis: census
6 ran · 0 no recorded run
valeoai/VideoActionModel community (classifier label)
classifier basis not recorded on the link
2 ran · 0 no recorded run
FunctionStatusWhere it lives
count_done_frames Ran
ran, honoured the contract we drafted
opendrivelab/driveagi/opendv/utils/frame_extraction.py
code served (permissive licence) · get_code("687ca21a8d19317f")
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 os
import tempfile

# Create a temporary directory with some files
temp_dir = tempfile.mkdtemp()
# Create a few dummy files to simulate frames
for i in range(5):
    open(os.path.join(temp_dir, f"frame_{i}.png"), 'w').close()

ARGS = {
    "save_path": temp_dir
}
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 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
get_mini_opendv Ran
ran; our contract draft was wrong, not the code
opendrivelab/driveagi/opendv/utils/download.py
code served (permissive licence) · get_code("01c31d25dffa247a")
How it ran

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

import torch
import numpy as np

torch.manual_seed(0)

# Construct a list of video info dictionaries
# The function filters for entries where subset == 'Mini'
full_video_list = [
    {
        'subset': 'Mini',
        'video_id': 'vid_001',
        'duration': 10.5,
        'frames': 250
    },
    {
        'subset': 'Train',
        'video_id': 'vid_002',
        'duration': 15.0,
        'frames': 360
    },
    {
        'subset': 'Mini',
        'video_id': 'vid_003',
        'duration': 8.2,
        'frames': 200
    },
    {
        'subset': 'Test',
        'video_id': 'vid_004',
        'duration': 12.0,
        'frames': 288
    },
]

ARGS = {
    'full_video_list': full_video_list
}
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 list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_video_with_meta Ran
ran; our contract draft was wrong, not the code
opendrivelab/driveagi/opendv/utils/download.py
code served (permissive licence) · get_code("ca3ca4be35ed28cc")
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 tempfile
import cv2
import numpy as np

# Create a temporary video file for testing
def create_test_video(video_path, fps=30, duration_sec=2, width=640, height=480):
    """Create a simple test video file"""
    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
    out = cv2.VideoWriter(video_path, fourcc, fps, (width, height))
    
    num_frames = int(fps * duration_sec)
    for i in range(num_frames):
        # Create a simple frame with changing color
        frame = np.zeros((height, width, 3), dtype=np.uint8)
        frame[:, :] = [i % 256, (i * 2) % 256, (i * 3) % 256]
        out.write(frame)
    
    out.release()

# Create a temporary directory and video file
temp_dir = tempfile.mkdtemp()
test_video_path = os.path.join(temp_dir, "test_video.mp4")
create_test_video(test_video_path, fps=30, duration_sec=2)

ARGS = {
    'video_path': test_video_path,
    'need_metas': ['fps', 'duration', 'num_frames']
}
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, 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
map_category_to_caption Ran
ran; our contract draft was wrong, not the code
opendrivelab/driveagi/opendv/utils/cmd2caption.py
code served (permissive licence) · get_code("cadbe4cde8ce2b48")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. 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 random

plain_caption_dict = {0: 'Go straight.', 1: 'Pass the intersection.', 2: 'Turn left.', 3: 'Turn right.', 4: 'Change to the left lane.', 5: 'Change to the right lane.', 6: 'Go to the left lane branch.', 7: 'Go to the right lane branch.', 8: 'Pass the crosswalk.', 9: 'Pass the railroad.', 10: 'Merge.', 11: 'Make a U-turn.', 12: 'Stop.', 13: 'Deviate.'}

diverse_caption_dict = {0: ['Move forward.', 'Move steady.', 'Go forward.', 'Go straight.', 'Proceed.', 'Drive forward.', 'Drive straight.', 'Drive steady.', 'Keep the direction.', 'Maintain the direction.'], 1: ['Pass the intersection.', 'Cross the intersection.', 'Traverse the intersection.', 'Drive through the intersection.', 'Move past the intersection.', 'Pass the junction.', 'Cross the junction.', 'Traverse the junction.', 'Drive through the junction.', 'Move past the junction.', 'Pass the crossroad.', 'Cross the crossroad.', 'Traverse the crossroad.', 'Drive through the crossroad.', 'Move past the crossroad.'], 2: ['Turn left.', 'Turn to the left.', 'Make a left turn.', 'Take a left turn.', 'Turn to the left.', 'Left turn.', 'Steer left.', 'Steer to the left.'], 3: ['Turn right.', 'Turn to the right.', 'Make a right turn.', 'Take a right turn.', 'Turn to the right.', 'Right turn.', 'Steer right.', 'Steer to the right.'], 4: ['Make a left lane change.', 'Change to the left lane.', 'Switch to the left lane.', 'Shift to the left lane.', 'Move to the left lane.'], 5: ['Make a right lane change.', 'Change to the right lane.', 'Switch to the right lane.', 'Shift to the right lane.', 'Move to the right lane.'], 6: ['Go to the left lane branch.', 'Take the left lane branch.', 'Move into the left lane branch.', 'Follow the left lane branch.', 'Follow the left side road.'], 7: ['Go to the right lane branch.', 'Take the right lane branch.', 'Move into the right lane branch.', 'Follow the right lane branch.', 'Follow the right side road.'], 8: ['Pass the crosswalk.', 'Cross the crosswalk.', 'Traverse the crosswalk.', 'Drive through the crosswalk.', 'Move past the crosswalk.', 'Pass the crossing area.', 'Cross the crossing area.', 'Traverse the crossing area.', 'Drive through the crossing area.', 'Move past the crossing area.'], 9: ['Pass the railroad.', 'Cross the railroad.', 'Traverse the railroad.', 'Drive through the railroad.', 'Move past the railroad.', 'Pass the railway.', 'Cross the railway.', 'Traverse the railway.', 'Drive through the railway.', 'Move past the railway.'], 10: ['Merge.', 'Merge traffic.', 'Merge into traffic.', 'Merge into the traffic.', 'Join the traffic.', 'Merge into the traffic flow.', 'Join the traffic flow.', 'Merge into the traffic stream.', 'Join the traffic stream.', 'Merge into the lane.'], 11: ['Make a U-turn.', 'Make a 180-degree turn.', 'Turn 180 degree.', 'Turn around.', 'Drive in a U-turn.'], 12: ['Stop.', 'Halt.', 'Decelerate.', 'Slow down.', 'Brake.'], 13: ['Deviate.', 'Deviate from the path.', 'Deviate from the lane.', 'Change the direction.', 'Shift the direction.']}

def map_category_to_caption(category_index, diverse=True):
    if diverse:
        return random.choice(diverse_caption_dict[category_index])
    else:
        return plain_caption_dict[category_index]

ARGS = {
    'category_index': 2,
    'diverse': True
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
mix_datasets Ran
ran on a synthesized input
valeoai/VideoActionModel/vam/datalib/data_mixing.py
code served (permissive licence) · get_code("2eb7a18c3d3c6298")
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 ConcatDataset. 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 random
from torch.utils.data import Dataset, ConcatDataset, Subset

torch.manual_seed(0)
random.seed(0)

class DummyDataset(Dataset):
    """Minimal dataset implementation for testing."""
    def __init__(self, size: int):
        self.size = size
    
    def __len__(self):
        return self.size
    
    def __getitem__(self, idx):
        return torch.randn(4), idx

# Create test datasets with distinct sizes
dataset1 = DummyDataset(10)
dataset2 = DummyDataset(8)
dataset3 = DummyDataset(12)

ARGS = {
    "datasets": [dataset1, dataset2, dataset3],
    "ratios": [0.3, 0.4, 0.2],
    "total_number_of_samples": 100,
    "seed": 42
}
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 ConcatDataset — a ConcatDataset 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
setup_logger Ran
ran on a synthesized input
valeoai/VideoActionModel/vam/datalib/create_opendv_tokens.py
code served (permissive licence) · get_code("7916970abec39bd5")
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 RootLogger. 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 logging
import os
import sys
import tempfile
from colorlog import ColoredFormatter

# Create a temporary directory for the logdir argument
temp_logdir = tempfile.mkdtemp()

# Define ARGS dictionary with all parameters for setup_logger
ARGS = {
    'logdir': temp_logdir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
colorlog — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type RootLogger — a RootLogger output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a RootLogger 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
special_video_setting_log Ran
ran, violated the contract we drafted
opendrivelab/driveagi/opendv/utils/frame_extraction.py
code served (permissive licence) · get_code("dc1a05c6429223c5")
How it ran

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

import json
import cv2
import tempfile
import os

# Create a mock video reader that has the get method
class MockVideoReader:
    def __init__(self, height=1080, width=1920):
        self._height = height
        self._width = width
    
    def get(self, prop):
        if prop == cv2.CAP_PROP_FRAME_HEIGHT:
            return self._height
        elif prop == cv2.CAP_PROP_FRAME_WIDTH:
            return self._width
        return 0

# Create a temporary file for exception logging
temp_exception_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.json')
exception_file_path = temp_exception_file.name
temp_exception_file.close()

# Create a mock video reader with 16:9 aspect ratio
mock_reader = MockVideoReader(height=720, width=1280)

ARGS = {
    'video_path': '/path/to/video.mp4',
    'exception_file': exception_file_path,
    'height': 720,
    'width': 1280,
    'video_reader': mock_reader
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2 — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
youtuber_formatize Ran
ran; our contract draft was wrong, not the code
opendrivelab/driveagi/opendv/utils/download.py
code served (permissive licence) · get_code("bd3e121688d8441b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. 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)

# Analyzing the function:
# youtuber_formatize(youtuber) calls youtuber.replace(' ', '_')
# This means youtuber must be a string with a .replace() method

ARGS = {
    "youtuber": "PewDiePie Gaming Channel"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 8 samples lifted from this paper's linked repositories, 8 carry a function identity and 0 do not. Of the 8 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 8 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

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

Repositories linked to this paper

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

Abstract

Figure 1. Overview of the GenAD paradigm. We aim to establish a generalized video prediction paradigm for autonomous driving by presenting the largest multimodal driving video dataset to date, OpenDV-2K, and a generative model that predicts the future given past visual and textual input, GenAD. The strong generalization and controllability of GenAD is validated spanning a diverse spectrum of tasks, including zero-shot domain transfer, language-conditioned prediction, action-conditioned prediction, and motion planning.

For agents

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

get_harvested_code_for_paper("2403.09630")
get_code_for_paper("2403.09630")
have("2403.09630")

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

curl https://syntology.ai/api/ran/2403.09630.json

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

[![Syntology run record](https://syntology.ai/api/ran/2403.09630.svg)](https://syntology.ai/paper/2403.09630)

Connect an agent — have() is free.