SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2501.01428 · 2025

GPT4Scene: Understand 3D Scenes from Videos with Vision-Language Models

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; 0 honoured the contract we drafted, 0 violated it and 8 have no contract verdict; 0 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/2501.01428.json · badge: /api/ran/2501.01428.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
Qi-Zhangyang/GPT4Scene canonical (classifier label)
classifier basis not recorded on the link
8 ran · 0 no recorded run
FunctionStatusWhere it lives
add_step_prompts Ran
ran on a synthesized input
Qi-Zhangyang/GPT4Scene/VLN-Ego-making/VLN-Ego/r2r_rxr_dataset_convert_sft.py
code served (permissive licence) · get_code("5064009b43a00726")
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 torch
import numpy as np
from tqdm import tqdm

torch.manual_seed(0)

# Create minimal navigation_data structure matching what the function expects
navigation_data = {
    'episode_1': {
        'instruction': 'Navigate to the kitchen and find the red object',
        'steps': [
            {
                'previous_images': ['image_0', 'image_1'],
                'current_image': 'image_2'
            },
            {
                'previous_images': ['image_1', 'image_2'],
                'current_image': 'image_3'
            }
        ]
    },
    'episode_2': {
        'instruction': 'Go to the bedroom',
        'steps': [
            {
                'previous_images': [],
                'current_image': 'image_0'
            },
            {
                'previous_images': ['image_0'],
                'current_image': 'image_1'
            }
        ]
    }
}

ARGS = {
    'navigation_data': navigation_data,
    'num_actions': 4
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch, tqdm — 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
bulid_message Ran
ran on a synthesized input
Qi-Zhangyang/GPT4Scene/evaluate/qwen2_vl.py
code served (permissive licence) · get_code("7173826fa5a3b3ec")
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 os
import tempfile
from PIL import Image

# Create a temporary image file for testing
temp_dir = tempfile.mkdtemp()
test_image_path = os.path.join(temp_dir, "test_image.jpg")

# Create a simple test image (200x150 pixels)
test_image = Image.new('RGB', (200, 150), color='red')
test_image.save(test_image_path)

# Create another test image for the list
test_image_path_2 = os.path.join(temp_dir, "test_image_2.jpg")
test_image_2 = Image.new('RGB', (300, 200), color='blue')
test_image_2.save(test_image_path_2)

# Create a minimal config class with the required 'resolution' attribute
class MinimalConfig:
    def __init__(self, resolution):
        self.resolution = resolution

# Build the arguments
ARGS = {
    'prompt': 'What is in this image?',
    'config': MinimalConfig(resolution=128),
    'eval_name': 'test_eval',
    'bev_mf_paths': [test_image_path, test_image_path_2]
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL — 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
create_dataset Ran
ran on a synthesized input
Qi-Zhangyang/GPT4Scene/evaluate/val_dataset.py
code served (permissive licence) · get_code("d1cc4570ba8bbd09")
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. 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 json
import tempfile
import os

torch.manual_seed(0)

# Create temporary annotation files
temp_dir = tempfile.mkdtemp()

# Create main annotation file
anno_data = [
    {
        'scene_id': 'scene_0',
        'obj_id': 1,
        'pred_id': 2,
        'sqa_type': 0,
        'prompt': 'What is in the scene?',
        'ref_captions': ['a car', 'a building'],
        'qid': 101
    },
    {
        'scene_id': 'scene_1',
        'obj_id': 3,
        'pred_id': 4,
        'eval_type': 1,
        'prompt': 'Describe the environment',
        'ref_captions': ['trees', 'road'],
        'qid': 102
    }
]

anno_file = os.path.join(temp_dir, 'anno.json')
with open(anno_file, 'w') as f:
    json.dump(anno_data, f)

# Create scene annotation file
scene_anno_data = {
    'scene_0': {'bev': '/path/to/bev_0', 'mf': '/path/to/mf_0'},
    'scene_1': {'bev': '/path/to/bev_1', 'mf': '/path/to/mf_1'}
}

scene_anno_file = os.path.join(temp_dir, 'scene_anno.json')
with open(scene_anno_file, 'w') as f:
    json.dump(scene_anno_data, f)

# Create a config class with all required attributes
class Config:
    def __init__(self):
        self.scene_anno = scene_anno_file
        self.val_tag = 'val_dataset'
        self.val_file_dict = {
            'val_dataset': anno_file
        }
        self.num_chunks = 2
        self.chunk_idx = 0

config = Config()

ARGS = {
    'config': 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 list — a list 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
create_loader Ran
ran on a synthesized input
Qi-Zhangyang/GPT4Scene/evaluate/val_dataset.py
code served (permissive licence) · get_code("c11a23c56c87bd2d")
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
from torch.utils.data import Dataset

torch.manual_seed(0)

class DummyDataset(Dataset):
    def __init__(self, size=2):
        self.size = size
    
    def __len__(self):
        return self.size
    
    def __getitem__(self, idx):
        prompt = f"prompt_{idx}"
        ref_caption = f"caption_{idx}"
        scene_id = f"scene_{idx}"
        bev_mf_path = f"path/to/bev_{idx}.npy"
        qid = f"q_{idx}"
        obj_id = f"obj_{idx}"
        pred_id = f"pred_{idx}"
        type_info = f"type_{idx}"
        
        return (prompt, ref_caption, scene_id, bev_mf_path, qid, obj_id, pred_id, type_info)

def custom_collate_fn(batch):
    prompts, ref_captions, scene_ids, bev_mf_paths, qids, obj_id, pred_ids, type_infos = zip(*batch)
    return {'custom_prompt': prompts, 'ref_captions': ref_captions, 'scene_id': scene_ids, 'bev_mf_paths': bev_mf_paths, 'qid': qids, 'obj_ids': obj_id, 'pred_ids': pred_ids, 'type_infos': type_infos}

dataset1 = DummyDataset(size=4)
dataset2 = DummyDataset(size=6)

ARGS = {
    'datasets': [dataset1, dataset2],
    'samplers': [None, None],
    'batch_size': [2, 3],
    'num_workers': [0, 0],
    'is_trains': [True, False],
    'collate_fns': [custom_collate_fn, custom_collate_fn]
}
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 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
create_navigation_data Ran
ran on a synthesized input
Qi-Zhangyang/GPT4Scene/VLN-Ego-making/VLN-Ego/r2r_rxr_dataset_convert_sft.py
code served (permissive licence) · get_code("13da697eeaf18b88")
How it ran

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

import os
import re
import json
import numpy as np
import tempfile
import shutil

# Create a temporary directory structure for testing
temp_dir = tempfile.mkdtemp()

# Create scene folder structure
scene_folder = os.path.join(temp_dir, "scene_001")
os.makedirs(scene_folder)

# Create trajectory subfolder
traj_folder = os.path.join(scene_folder, "trajectory_001")
os.makedirs(traj_folder)

# Create instruction.txt
with open(os.path.join(traj_folder, "instruction.txt"), "w") as f:
    f.write("Navigate to the kitchen")

# Create action_list.json with 8 actions
actions = [1, 2, 3, 0, 1, 2, 1, 0]
with open(os.path.join(traj_folder, "action_list.json"), "w") as f:
    json.dump(actions, f)

# Create dummy step images
for i in range(len(actions)):
    step_image_path = os.path.join(traj_folder, f"step_{i}.jpg")
    with open(step_image_path, "w") as f:
        f.write(f"dummy image data {i}")

# Define ARGS
ARGS = {
    "base_folder": temp_dir,
    "max_frames": 4,
    "num_actions": 2
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tqdm — 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
create_sampler Ran
ran on a synthesized input
Qi-Zhangyang/GPT4Scene/evaluate/val_dataset.py
code served (permissive licence) · get_code("8d7bf7ad0e60457f")
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) 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
from torch.utils.data import Dataset

torch.manual_seed(0)

class DummyDataset(Dataset):
    def __init__(self, size=10):
        self.size = size
    
    def __len__(self):
        return self.size
    
    def __getitem__(self, idx):
        return torch.randn(4), torch.randint(0, 2, (1,))

dataset1 = DummyDataset(size=10)
dataset2 = DummyDataset(size=8)

ARGS = {
    "datasets": [dataset1, dataset2],
    "shuffles": [True, False],
    "num_tasks": 2,
    "global_rank": 0
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
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_step_images Ran
ran on a synthesized input
Qi-Zhangyang/GPT4Scene/VLN-Ego-making/VLN-Ego/r2r_rxr_dataset_convert_sft.py
code served (permissive licence) · get_code("10db9373c4cbf700")
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 os
import tempfile
import shutil

# Create a temporary directory with step image files
temp_dir = tempfile.mkdtemp()

# Create some dummy step image files
step_files = ['step_0.jpg', 'step_1.jpg', 'step_10.jpg', 'step_2.jpg', 'step_5.jpg']
for filename in step_files:
    filepath = os.path.join(temp_dir, filename)
    with open(filepath, 'w') as f:
        f.write('dummy image data')

# Also add some non-matching files to test filtering
with open(os.path.join(temp_dir, 'other_file.jpg'), 'w') as f:
    f.write('not a step file')
with open(os.path.join(temp_dir, 'step_invalid.jpg'), 'w') as f:
    f.write('invalid step file')

ARGS = {
    'ep_folder': 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 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
preprocess_image_nearest Ran
ran on a synthesized input
Qi-Zhangyang/GPT4Scene/evaluate/qwen2_vl.py
code served (permissive licence) · get_code("174c802b9d2bb222")
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 int; 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 tempfile
from PIL import Image
import os

# Create a temporary test image
temp_dir = tempfile.mkdtemp()
test_image_path = os.path.join(temp_dir, "test_image.png")

# Create a simple RGB image with specific dimensions
test_image = Image.new('RGB', (800, 600), color='red')
test_image.save(test_image_path)

ARGS = {
    'image_path': test_image_path,
    'image_resolution': 512,
    'perform_resize': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL — 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 int — 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

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, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 0 later) and 7 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
create_sampler 1 other paper
1 dated earlier
2305.07922 (2023-05)
CodeT5+: Open Code Large Language Models for Code Understanding and Generation

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

Repositories linked to this paper

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

Abstract

In recent years, 2D Vision-Language Models (VLMs) have made significant strides in image-text understanding tasks. However, their performance in 3D spatial comprehension, which is critical for embodied intelligence, remains limited. Recent advances have leveraged 3D point clouds and multi-view images as inputs, yielding promising results. However, we propose exploring a purely vision-based solution inspired by human perception, which merely relies on visual cues for 3D spatial understanding. This paper empirically investigates the limitations of VLMs in 3D spatial knowledge, revealing that their primary shortcoming lies in the lack of global-local correspondence between the scene and individual frames. To address this, we introduce GPT4Scene, a novel visual prompting paradigm in VLM training and inference that helps build the global-local relationship, significantly improving the 3D spatial understanding of indoor scenes. Specifically, GPT4Scene constructs a Bird's Eye View (BEV) image from the video and marks consistent object IDs across both frames and the BEV image. The model then inputs the concatenated BEV image and video frames with markers. In zero-shot evaluations, GPT4Scene improves performance over closed-source VLMs like GPT-4o. Additionally, we prepare a processed video dataset consisting of 165K text annotation to fine-tune open-source VLMs, achieving state-of-the-art performance on all 3D understanding tasks. Surprisingly, after training with the GPT4Scene paradigm, VLMs consistently improve during inference, even without object marker prompting and BEV image as explicit correspondence. It demonstrates that the proposed paradigm helps VLMs develop an intrinsic ability to understand 3D scenes, which paves the way for a seamless approach to extending pre-trained VLMs for 3D scene understanding.

For agents

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

get_harvested_code_for_paper("2501.01428")
get_code_for_paper("2501.01428")
have("2501.01428")

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

curl https://syntology.ai/api/ran/2501.01428.json

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

[![Syntology run record](https://syntology.ai/api/ran/2501.01428.svg)](https://syntology.ai/paper/2501.01428)

Connect an agent — have() is free.