SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2409.18124 · 2024

Lotus: Diffusion-based Visual Foundation Model for High-quality Dense Prediction

arXiv · PDF · Open in the Atlas

Code that ran

13 of 14 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, 2 violated it and 7 have no contract verdict; 4 failures of our instrument; 1 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/2409.18124.json · badge: /api/ran/2409.18124.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
envision-research/lotus community (classifier label)
classifier basis not recorded on the link
13 ran · 1 no recorded run
FunctionStatusWhere it lives
align_normals Ran
ran on a synthesized input
envision-research/lotus/utils/hypersim_dataset.py
code served (permissive licence) · get_code("fb3ab3bf8b931d62")
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 3, dtype family float, shape (4, 3, 3) 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

# Create small valid arguments for align_normals
H = 4
W = 3

# K should contain focal lengths and principal point: [fx, fy, cx, cy]
K = np.array([500.0, 500.0, 1.5, 2.0], dtype=np.float64)

# depth should be H x W array of depth values
depth = np.ones((H, W), dtype=np.float64) * 2.0

# normal should be H x W x 3 array of surface normals
normal = np.random.randn(H, W, 3).astype(np.float64)
# Normalize the normals
normal = normal / (np.linalg.norm(normal, axis=2, keepdims=True) + 1e-8)

ARGS = {
    "normal": normal,
    "depth": depth,
    "K": K,
    "H": H,
    "W": W
}
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 ndarray, ndim 3, dtype family float, shape (4, 3, 3) — shape (4, 3, 3) 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
colorize_depth_map Ran
ran on a synthesized input
envision-research/lotus/utils/image_utils.py
code served (permissive licence) · get_code("03c4ff24ca415b85")
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 Image. 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

torch.manual_seed(0)

# Create a small depth map (2D numpy array)
depth = np.random.rand(4, 3).astype(np.float32)

# Create a mask as a torch tensor (boolean mask with same spatial dims)
mask = torch.tensor([[True, False, True],
                     [False, True, False],
                     [True, True, False],
                     [False, False, True]])

# Define arguments
ARGS = {
    'depth': depth,
    'mask': mask,
    'reverse_color': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, matplotlib, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Image — a Image output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Image output the recorded ndim 3 and dtype family int 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
copy_images_and_depths Ran
ran on a synthesized input
envision-research/lotus/utils/process_hypersim.py
code served (permissive licence) · get_code("46450d32e73bc5a8")
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 shutil
import h5py
import numpy as np
import pandas as pd
import tempfile

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

# Create sample HDF5 files
def create_sample_hdf5(path, has_nan_values=False):
    os.makedirs(os.path.dirname(path), exist_ok=True)
    with h5py.File(path, 'w') as f:
        if has_nan_values:
            data = np.array([[1.0, np.nan], [3.0, 4.0]])
        else:
            data = np.array([[1.0, 2.0], [3.0, 4.0]])
        f.create_dataset('depth', data=data)

# Create sample image file
def create_sample_image(path):
    os.makedirs(os.path.dirname(path), exist_ok=True)
    with open(path, 'wb') as f:
        f.write(b'fake image data')

# Set up test data structure
scene_name = "test_scene"
camera_name = "camera_01"

# Create sample rows for dataframe
rows = []
for frame_id in range(2):
    rows.append({
        'frame_id': frame_id,
        'scene_name': scene_name,
        'camera_name': camera_name
    })

df = pd.DataFrame(rows)

# Create actual files in source directory
for frame_id in range(2):
    image_file = f"frame.{frame_id:04d}.tonemap.jpg"
    depth_file = f"frame.{frame_id:04d}.depth_meters.hdf5"
    normal_file = f"frame.{frame_id:04d}.normal_cam.hdf5"
    
    src_image_path = os.path.join(temp_src, scene_name, 'images', f"scene_{camera_name}_final_preview", image_file)
    src_depth_path = os.path.join(temp_src, scene_name, 'images', f"scene_{camera_name}_geometry_hdf5", depth_file)
    src_normal_path = os.path.join(temp_src, scene_name, 'images', f"scene_{camera_name}_geometry_hdf5", normal_file)
    
    create_sample_image(src_image_path)
    create_sample_hdf5(src_depth_path, has_nan_values=False)
    create_sample_hdf5(src_normal_path, has_nan_values=False)

ARGS = {
    'df': df,
    'src_path': temp_src,
    'trg_path': temp_trg,
    'filter_nan': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
h5py, numpy, pandas, tqdm — 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
count_data Ran
ran on a synthesized input
envision-research/lotus/utils/depth2normal.py
code served (permissive licence) · get_code("54d4f5d89dd5b02c")
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
import shutil

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

# Create mock directory structure with depth files
for scene_num in [1, 2, 3]:
    scene_dir = os.path.join(temp_dir, f'Scene{scene_num}', 'images')
    os.makedirs(scene_dir, exist_ok=True)
    
    # Create some depth files
    for i in range(3):
        depth_file = os.path.join(scene_dir, f'depth_{i}.png')
        with open(depth_file, 'w') as f:
            f.write('mock depth data')
    
    # Create some non-depth files
    for i in range(2):
        other_file = os.path.join(scene_dir, f'rgb_{i}.png')
        with open(other_file, 'w') as f:
            f.write('mock rgb data')

# Define the arguments
ARGS = {
    'root_dir': temp_dir,
    'scenes': [1, 2, 3]
}
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
creat_uv_mesh Ran
ran on a synthesized input
envision-research/lotus/utils/hypersim_dataset.py
code served (permissive licence) · get_code("591897bcb1b5bd2b")
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 2, dtype family float, shape (3, 12) 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

# The function creat_uv_mesh takes two parameters: H (height) and W (width)
# It creates a UV mesh by:
# 1. Creating a meshgrid of coordinates for H rows and W columns
# 2. Stacking x and y coordinates
# 3. Adding a row of ones
# 4. Concatenating them together

# Both H and W should be positive integers representing dimensions

ARGS = {
    'H': 4,
    'W': 3
}
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 ndarray, ndim 2, dtype family float, shape (3, 12) — shape (3, 12) 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
has_nan Ran
ran on a synthesized input
envision-research/lotus/utils/process_hypersim.py
code served (permissive licence) · get_code("c8e80fa247270366")
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 h5py
import numpy as np
import tempfile
import os

# Create a temporary HDF5 file with some test data
temp_dir = tempfile.mkdtemp()
hdf5_path = os.path.join(temp_dir, "test.h5")

with h5py.File(hdf5_path, 'w') as f:
    f.create_dataset('data1', data=np.array([1.0, 2.0, 3.0, 4.0]))
    f.create_dataset('data2', data=np.array([5.0, 6.0, 7.0, 8.0]))

ARGS = {
    'hdf5_path': hdf5_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
h5py, numpy — 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
intrins_crop Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
envision-research/lotus/utils/projection.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from VisualComputingInstitute/diffusion-e2e-ft, is pointer only (licence: NONE) · get_code("07d8c13c1190a20a")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from VisualComputingInstitute/diffusion-e2e-ft, 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.

intrins_to_zero Ran
ran, violated the contract we drafted
envision-research/lotus/utils/projection.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from VisualComputingInstitute/diffusion-e2e-ft, is pointer only (licence: NONE) · get_code("1218c2650fe6ffed")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from VisualComputingInstitute/diffusion-e2e-ft, 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.

intrins_zero_to Ran
ran, violated the contract we drafted
envision-research/lotus/utils/projection.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from VisualComputingInstitute/diffusion-e2e-ft, is pointer only (licence: NONE) · get_code("f56119d671b23caa")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from VisualComputingInstitute/diffusion-e2e-ft, 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.

rescale_noise_cfg Ran
ran; our contract draft was wrong, not the code
envision-research/lotus/pipeline.py
code served (permissive licence) · get_code("bea2d776a332f2b0")
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 Tensor, ndim 4, dtype family float, shape (2, 4, 3, 3) 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 small tensors with distinct dimensions
# noise_cfg: shape (batch=2, channels=4, height=3, width=3)
noise_cfg = torch.randn(2, 4, 3, 3)

# noise_pred_text: shape (batch=2, channels=4, height=3, width=3)
noise_pred_text = torch.randn(2, 4, 3, 3)

# guidance_rescale: float between 0 and 1
guidance_rescale = 0.7

ARGS = {
    "noise_cfg": noise_cfg,
    "noise_pred_text": noise_pred_text,
    "guidance_rescale": guidance_rescale,
}
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 Tensor, ndim 4, dtype family float, shape (2, 4, 3, 3) — shape (2, 4, 3, 3) 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
resize_back Ran
ran on a synthesized input
envision-research/lotus/utils/image_utils.py
code served (permissive licence) · get_code("56b4266634154e42")
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, 8, 6) 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
import numpy as np
import PIL.Image
from torchvision.transforms import InterpolationMode

torch.manual_seed(0)

# Create a small torch tensor (batch=2, channels=3, height=4, width=5)
tensor_img = torch.randn(2, 3, 4, 5)

# Create a small numpy array (batch=2, height=4, width=5, channels=3)
numpy_img = np.random.randn(2, 4, 5, 3).astype(np.float32)

# Create small PIL images
pil_img = PIL.Image.new('RGB', (5, 4))
pil_img_list = [PIL.Image.new('RGB', (5, 4)) for _ in range(2)]

# Target size as tuple
target_size = (8, 6)

# Resample method
resample_method = InterpolationMode.BILINEAR

ARGS = {
    "img": tensor_img,
    "target_size": target_size,
    "resample_method": resample_method
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy, torch, torchvision — 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, 8, 6) — shape (2, 3, 8, 6) 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
resize_max_res Ran
ran; our contract draft was wrong, not the code
envision-research/lotus/utils/image_utils.py
code served (permissive licence) · get_code("715b48a5e7426c5e")
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, 3) 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
from torchvision.transforms import InterpolationMode

torch.manual_seed(0)

# Create a small batch of images with shape [B, C, H, W]
# B=2 (batch), C=3 (RGB), H=8, W=6
img = torch.randint(0, 256, (2, 3, 8, 6), dtype=torch.uint8).float() / 255.0

# Maximum edge resolution
max_edge_resolution = 4

# Resampling method
resample_method = InterpolationMode.BILINEAR

ARGS = {
    "img": img,
    "max_edge_resolution": max_edge_resolution,
    "resample_method": resample_method,
}
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 Tensor, ndim 4, dtype family float, shape (2, 3, 4, 3) — shape (2, 3, 4, 3) 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
retrieve_timesteps Ran
ran; our contract draft was wrong, not the code
envision-research/lotus/pipeline.py
code served (permissive licence) · get_code("6cccd001a059f194")
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 inspect
from typing import List, Optional, Union

torch.manual_seed(0)

# Create a minimal scheduler class that satisfies the function's needs
class MockScheduler:
    def __init__(self, supports_timesteps=True):
        self.supports_timesteps = supports_timesteps
        self.timesteps = torch.tensor([999, 500, 100, 10])
    
    def set_timesteps(self, num_inference_steps=None, device=None, timesteps=None, **kwargs):
        if timesteps is not None:
            self.timesteps = torch.tensor(timesteps, device=device)
        else:
            # Simulate creating timesteps based on num_inference_steps
            self.timesteps = torch.linspace(999, 0, num_inference_steps, dtype=torch.long, device=device)

# Create scheduler instance
scheduler = MockScheduler(supports_timesteps=True)

# Define arguments for the retrieve_timesteps function
ARGS = {
    "scheduler": scheduler,
    "num_inference_steps": 4,
    "device": "cpu",
    "timesteps": None,
}
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 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
hypersim_distance_to_depth Not yet run
unverified (harvested, no recorded run)
envision-research/lotus/utils/hypersim_dataset.py
code served (permissive licence) · get_code("6b2450686c042fff")
How it ran

no invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 distinct functions with an identity, 6 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 5 the same month, 2 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.

FunctionAlso present inEarliest-datedOthers named
rescale_noise_cfg 62 other papers
33 dated earlier, 4 the same month, 24 later, 1 same year, order not determinable
2206.02779 (2022-06)
Blended Latent Diffusion
2304.06720 (2023-04)
Expressive Text-to-Image Generation with Rich Text
2309.06380 (2023-09)
InstaFlow: One Step is Enough for High-Quality Diffusion-Based Text-to-Image Generation
and 59 more not named here
retrieve_timesteps 21 other papers
11 dated earlier, 2 the same month, 8 later
2310.06313 (2023-10)
Advancing Pose-Guided Image Synthesis with Progressive Conditional Diffusion Models
2401.11708 (2024-01)
Mastering Text-to-Image Diffusion: Recaptioning, Planning, and Generating with Multimodal LLMs
2403.06976 (2024-03)
BrushNet: A Plug-and-Play Image Inpainting Model with Decomposed Dual-Branch Diffusion
and 18 more not named here
intrins_crop 1 other paper
1 the same month
2409.11355 (2024-09)
Fine-Tuning Image-Conditional Diffusion Models is Easier than You Think
intrins_to_zero 1 other paper
1 the same month
2409.11355 (2024-09)
Fine-Tuning Image-Conditional Diffusion Models is Easier than You Think
intrins_zero_to 1 other paper
1 the same month
2409.11355 (2024-09)
Fine-Tuning Image-Conditional Diffusion Models is Easier than You Think
resize_max_res 1 other paper
1 dated earlier
2312.02145 (2023-12)
Repurposing Diffusion-Based Image Generators for Monocular Depth Estimation

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

Leveraging the visual priors of pre-trained text-to-image diffusion models offers a promising solution to enhance zero-shot generalization in dense prediction tasks. However, existing methods often uncritically use the original diffusion formulation, which may not be optimal due to the fundamental differences between dense prediction and image generation. In this paper, we provide a systemic analysis of the diffusion formulation for the dense prediction, focusing on both quality and efficiency. And we find that the original parameterization type for image generation, which learns to predict noise, is harmful for dense prediction; the multi-step noising/denoising diffusion process is also unnecessary and challenging to optimize. Based on these insights, we introduce Lotus, a diffusion-based visual foundation model with a simple yet effective adaptation protocol for dense prediction. Specifically, Lotus is trained to directly predict annotations instead of noise, thereby avoiding harmful variance. We also reformulate the diffusion process into a single-step procedure, simplifying optimization and significantly boosting inference speed. Additionally, we introduce a novel tuning strategy called detail preserver, which achieves more accurate and fine-grained predictions. Without scaling up the training data or model capacity, Lotus achieves SoTA performance in zero-shot depth and normal estimation across various datasets. It also enhances efficiency, being significantly faster than most existing diffusion-based methods. Lotus' superior quality and efficiency also enable a wide range of practical applications, such as joint estimation, single/multi-view 3D reconstruction, etc. Project page: https://lotus3d.github.io/.

For agents

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

get_harvested_code_for_paper("2409.18124")
get_code_for_paper("2409.18124")
have("2409.18124")

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

curl https://syntology.ai/api/ran/2409.18124.json

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

[![Syntology run record](https://syntology.ai/api/ran/2409.18124.svg)](https://syntology.ai/paper/2409.18124)

Connect an agent — have() is free.