SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2412.04463 · 2024

MegaSaM: Accurate, Fast, and Robust Structure and Motion from Casual Dynamic Videos

arXiv · PDF · Open in the Atlas

Code that ran

12 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; 2 honoured the contract we drafted, 0 violated it and 6 have no contract verdict; 4 failures of our instrument; 2 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/2412.04463.json · badge: /api/ran/2412.04463.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
mega-sam/mega-sam community (classifier label)
classifier basis not recorded on the link
12 ran · 2 no recorded run
FunctionStatusWhere it lives
get_camera_rays Ran
ran on a synthesized input
mega-sam/mega-sam/cvd_opt/geometry_utils.py
code served (permissive licence) · get_code("bda1ea37848c1a5f")
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 3, dtype family float, shape (2, 3, 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)

# Define small tensors for the function
batch_size = 2
num_points = 4

# world_T_cam_b44: batch of 4x4 transformation matrices (world to camera)
world_T_cam_b44 = torch.eye(4).unsqueeze(0).repeat(batch_size, 1, 1)
world_T_cam_b44[0, 0:3, 3] = torch.tensor([1.0, 2.0, 3.0])
world_T_cam_b44[1, 0:3, 3] = torch.tensor([0.5, 1.5, 2.5])

# world_points_b3N: batch of 3xN point coordinates in world frame
world_points_b3N = torch.randn(batch_size, 3, num_points)

# cam_T_world_b44: batch of 4x4 transformation matrices (camera to world)
# This is the inverse of world_T_cam_b44
cam_T_world_b44 = torch.inverse(world_T_cam_b44)

# Test both cases
ARGS_case_1 = {
    'world_T_cam_b44': world_T_cam_b44,
    'world_points_b3N': world_points_b3N,
    'in_camera_frame': False,
    'cam_T_world_b44': None,
    'eps': 0.0001
}

ARGS_case_2 = {
    'world_T_cam_b44': world_T_cam_b44,
    'world_points_b3N': world_points_b3N,
    'in_camera_frame': True,
    'cam_T_world_b44': cam_T_world_b44,
    'eps': 0.0001
}

# Use the case where in_camera_frame=False as the default
ARGS = ARGS_case_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 Tensor, ndim 3, dtype family float, shape (2, 3, 4) — shape (2, 3, 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
gradient_loss Ran
ran on a synthesized input
mega-sam/mega-sam/cvd_opt/cvd_opt.py
code served (permissive licence) · get_code("de454ae3ade5d9ae")
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 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 torch

torch.manual_seed(0)

# The function expects:
# - gt: a tensor with spatial dimensions (height, width)
# - pred: a tensor with spatial dimensions (height, width)
# - u: unused parameter
# The tensors need to be at least 4x4 in spatial dimensions to allow slicing operations
# like [..., 0:-2, 1:-1], [..., 2:, 1:-1], etc.

# Create ground truth tensor: batch_size=2, channels=1, height=6, width=6
gt = torch.randn(2, 1, 6, 6)

# Create prediction tensor with same shape
pred = torch.randn(2, 1, 6, 6)

# u is unused in the function
u = None

ARGS = {
    "gt": gt,
    "pred": pred,
    "u": u,
}
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 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
pose_distance Ran
ran on a synthesized input
mega-sam/mega-sam/cvd_opt/geometry_utils.py
code served (permissive licence) · get_code("1e7d6a056f2488fa")
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 2, 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

torch.manual_seed(0)

# Create a batch of 2 pose matrices (4x4 homogeneous transformation matrices)
# Each pose_b44[i] is a 4x4 matrix with rotation in [:3, :3] and translation in [:3, 3]
pose_b44 = torch.tensor([
    [
        [0.9, 0.1, 0.0, 1.5],
        [-0.1, 0.9, 0.0, 2.0],
        [0.0, 0.0, 1.0, 0.5],
        [0.0, 0.0, 0.0, 1.0]
    ],
    [
        [0.8, 0.2, 0.0, -1.0],
        [-0.2, 0.8, 0.0, 1.0],
        [0.0, 0.0, 1.0, -0.5],
        [0.0, 0.0, 0.0, 1.0]
    ]
], dtype=torch.float32)

ARGS = {
    "pose_b44": pose_b44
}
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, ndim 2, 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
read_cameras_binary Ran
ran; our contract draft was wrong, not the code
mega-sam/mega-sam/colmap_read_model.py
code served (permissive licence) · get_code("1f7b4cd3dea533ca")
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 tempfile
import struct
import os
import collections

CameraModel = collections.namedtuple('CameraModel', ['model_id', 'model_name', 'num_params'])

CAMERA_MODELS = {
    CameraModel(model_id=0, model_name='SIMPLE_PINHOLE', num_params=3),
    CameraModel(model_id=1, model_name='PINHOLE', num_params=4),
    CameraModel(model_id=2, model_name='SIMPLE_RADIAL', num_params=4),
}

CAMERA_MODEL_IDS = dict([(camera_model.model_id, camera_model) for camera_model in CAMERA_MODELS])

# Create a temporary binary file with valid camera data
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.bin')
temp_path = temp_file.name

try:
    # Write number of cameras (1 camera)
    num_cameras = 1
    temp_file.write(struct.pack('<Q', num_cameras))
    
    # Write camera properties: camera_id, model_id, width, height
    camera_id = 1
    model_id = 1  # PINHOLE model
    width = 640
    height = 480
    temp_file.write(struct.pack('<iiQQ', camera_id, model_id, width, height))
    
    # Write camera parameters (4 parameters for PINHOLE model)
    params = [500.0, 500.0, 320.0, 240.0]
    for param in params:
        temp_file.write(struct.pack('<d', param))
    
    temp_file.close()
    
    ARGS = {
        'path_to_model_file': temp_path
    }

finally:
    pass
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 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
read_cameras_binary Ran
ran; our contract draft was wrong, not the code
mega-sam/mega-sam/camera_tracking_scripts/colmap_read_model.py
code served (permissive licence) · get_code("c9dc32524dba1c3e")
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 collections
import struct
import numpy as np
import tempfile
import os

CameraModel = collections.namedtuple('CameraModel', ['model_id', 'model_name', 'num_params'])
Camera = collections.namedtuple('Camera', ['id', 'model', 'width', 'height', 'params'])

CAMERA_MODELS = {CameraModel(model_id=0, model_name='SIMPLE_PINHOLE', num_params=3), CameraModel(model_id=1, model_name='PINHOLE', num_params=4), CameraModel(model_id=2, model_name='SIMPLE_RADIAL', num_params=4), CameraModel(model_id=3, model_name='RADIAL', num_params=5), CameraModel(model_id=4, model_name='OPENCV', num_params=8), CameraModel(model_id=5, model_name='OPENCV_FISHEYE', num_params=8), CameraModel(model_id=6, model_name='FULL_OPENCV', num_params=12), CameraModel(model_id=7, model_name='FOV', num_params=5), CameraModel(model_id=8, model_name='SIMPLE_RADIAL_FISHEYE', num_params=4), CameraModel(model_id=9, model_name='RADIAL_FISHEYE', num_params=5), CameraModel(model_id=10, model_name='THIN_PRISM_FISHEYE', num_params=12)}

CAMERA_MODEL_IDS = dict([(camera_model.model_id, camera_model) for camera_model in CAMERA_MODELS])

def create_test_cameras_binary_file():
    """Create a minimal valid binary cameras file for testing."""
    temp_file = tempfile.NamedTemporaryFile(mode='wb', delete=False, suffix='.bin')
    temp_path = temp_file.name
    
    # Write number of cameras (2 cameras)
    num_cameras = 2
    temp_file.write(struct.pack('<Q', num_cameras))
    
    # Camera 1: PINHOLE model (model_id=1, num_params=4)
    camera_id_1 = 1
    model_id_1 = 1
    width_1 = 640
    height_1 = 480
    params_1 = [500.0, 500.0, 320.0, 240.0]  # fx, fy, cx, cy
    
    temp_file.write(struct.pack('<iiQQ', camera_id_1, model_id_1, width_1, height_1))
    temp_file.write(struct.pack('<' + 'd' * 4, *params_1))
    
    # Camera 2: RADIAL model (model_id=3, num_params=5)
    camera_id_2 = 2
    model_id_2 = 3
    width_2 = 1280
    height_2 = 960
    params_2 = [600.0, 600.0, 640.0, 480.0, 0.1]  # fx, fy, cx, cy, k1
    
    temp_file.write(struct.pack('<iiQQ', camera_id_2, model_id_2, width_2, height_2))
    temp_file.write(struct.pack('<' + 'd' * 5, *params_2))
    
    temp_file.close()
    return temp_path

binary_file_path = create_test_cameras_binary_file()

ARGS = {
    'path_to_model_file': binary_file_path
}
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 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
read_cameras_text Ran
ran; our contract draft was wrong, not the code
mega-sam/mega-sam/colmap_read_model.py
code served (permissive licence) · get_code("b768293892633d35")
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 collections
import numpy as np
import tempfile
import os

Camera = collections.namedtuple('Camera', ['id', 'model', 'width', 'height', 'params'])

# Create a temporary file with valid camera data
temp_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt')
temp_file.write("# Camera list\n")
temp_file.write("1 SIMPLE_PINHOLE 640 480 320.5 240.5 500.0\n")
temp_file.write("2 PINHOLE 1280 960 640.0 480.0 1000.0 0.5\n")
temp_file.write("3 SIMPLE_RADIAL 800 600 400.0 300.0 600.0 0.1\n")
temp_file.close()

ARGS = {
    'path': temp_file.name
}
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 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
read_cameras_text Ran
ran; our contract draft was wrong, not the code
mega-sam/mega-sam/camera_tracking_scripts/colmap_read_model.py
code served (permissive licence) · get_code("35498794a9af2019")
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 tempfile
import os
import collections
import numpy as np

# Create a temporary file with camera data in COLMAP format
temp_dir = tempfile.mkdtemp()
cameras_file = os.path.join(temp_dir, 'cameras.txt')

# Write a sample cameras.txt file
with open(cameras_file, 'w') as f:
    f.write("# Camera list with one line of data per camera:\n")
    f.write("# CAMERA_ID, MODEL, WIDTH, HEIGHT, PARAMS[]\n")
    f.write("# Number of cameras: 2\n")
    f.write("1 SIMPLE_PINHOLE 640 480 320.5 240.5 500.0\n")
    f.write("2 PINHOLE 1280 960 640.0 480.0 1000.0 0.5\n")

ARGS = {
    'path': cameras_file
}
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 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
read_next_bytes Ran
ran, honoured the contract we drafted
mega-sam/mega-sam/colmap_read_model.py
code served (permissive licence) · get_code("735c9f07e56255b5")
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 struct
import io

# Create a binary file-like object with test data
# We'll write some bytes that can be unpacked according to a format string
test_data = struct.pack('<hHiI', 100, 200, 300, 400)
fid = io.BytesIO(test_data)

ARGS = {
    'fid': fid,
    'num_bytes': 12,
    'format_char_sequence': 'hHiI',
    'endian_character': '<'
}
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 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
read_next_bytes Ran
ran, honoured the contract we drafted
mega-sam/mega-sam/camera_tracking_scripts/colmap_read_model.py
code served (permissive licence) · get_code("f651b415cb3682d6")
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 struct
import io

# Create a binary file-like object with test data
# We'll pack some test values and create a file object to read from
test_data = struct.pack('<hf', 42, 3.14)  # short (2 bytes) + float (4 bytes)
fid = io.BytesIO(test_data)

ARGS = {
    'fid': fid,
    'num_bytes': 6,
    'format_char_sequence': 'hf',
    'endian_character': '<'
}
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 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
resize_flow Ran
ran on a synthesized input
mega-sam/mega-sam/cvd_opt/preprocess_flow.py
code served (permissive licence) · get_code("72464163e1ed4a8b")
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 (8, 10, 2) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import cv2
import numpy as np

# Create a small optical flow array
# flow has shape (height, width, 2) where last dimension is (u, v) components
flow_h, flow_w = 4, 5
flow = np.random.randn(flow_h, flow_w, 2).astype(np.float32)

# Target image dimensions
img_h, img_w = 8, 10

ARGS = {
    'flow': flow,
    'img_h': img_h,
    'img_w': img_w
}
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 ndarray, ndim 3, dtype family float, shape (8, 10, 2) — shape (8, 10, 2) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
si_loss Ran
ran on a synthesized input
mega-sam/mega-sam/cvd_opt/cvd_opt.py
code served (permissive licence) · get_code("ac04d341cfd51f93")
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 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 torch

torch.manual_seed(0)

# The function si_loss expects:
# - gt: a tensor with shape (batch, height, width) that gets viewed as (batch, -1)
# - pred: a tensor with shape (batch, height, width) that gets viewed as (batch, -1)
# Both tensors are clamped, logged, and compared element-wise

# Create small tensors with batch size 2 and spatial dimensions 4x3
batch_size = 2
height = 4
width = 3

# Ground truth tensor: values in range [0.1, 100] to be safely within clamp range
gt = torch.rand(batch_size, height, width) * 99.9 + 0.1

# Prediction tensor: similar range
pred = torch.rand(batch_size, height, width) * 99.9 + 0.1

ARGS = {
    "gt": gt,
    "pred": pred,
}
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 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
warp_flow Ran
ran on a synthesized input
mega-sam/mega-sam/cvd_opt/preprocess_flow.py
code served (permissive licence) · get_code("b0ed5fdc9c9400a1")
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 int, shape (4, 5, 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
import cv2

# Create a small image (3 channels, 4 rows, 5 columns)
img = np.random.randint(0, 256, size=(4, 5, 3), dtype=np.uint8)

# Create optical flow field (same spatial dimensions as image, 2 channels for x,y displacement)
flow = np.random.randn(4, 5, 2).astype(np.float32) * 0.5

ARGS = {
    "img": img,
    "flow": flow,
}
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 ndarray, ndim 3, dtype family int, shape (4, 5, 3) — shape (4, 5, 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
sobel_fg_alpha Not yet run
unverified (harvested, no recorded run)
mega-sam/mega-sam/cvd_opt/cvd_opt.py
code served (permissive licence) · get_code("7e492e661d45f956")
How it ran

no invocation recorded.

to_homogeneous Not yet run
unverified (harvested, no recorded run)
mega-sam/mega-sam/cvd_opt/geometry_utils.py
code served (permissive licence) · get_code("71da5d76a02a0fe3")
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 12 distinct functions with an identity, 4 are also present in other papers whose code we hold (4 with at least one paper dated earlier, 1 the same month, 4 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
read_next_bytes 70 other papers
62 dated earlier, 2 the same month, 6 later
1804.00650 (2018-04)
DeepMVS: Learning Multi-view Stereopsis
1902.10556 (2019-02)
Recurrent MVSNet for High-resolution Multi-view Stereo Depth Inference
1912.06378 (2019-12)
Cascade Cost Volume for High-Resolution Multi-View Stereo and Stereo Matching
and 67 more not named here
read_cameras_text 37 other papers
34 dated earlier, 3 later
1902.10556 (2019-02)
Recurrent MVSNet for High-resolution Multi-view Stereo Depth Inference
1912.06378 (2019-12)
Cascade Cost Volume for High-Resolution Multi-View Stereo and Stereo Matching
1912.11744 (2019-12)
Planar Prior Assisted PatchMatch Multi-View Stereo
and 34 more not named here
read_cameras_binary 21 other papers
20 dated earlier, 1 later
1902.10556 (2019-02)
Recurrent MVSNet for High-resolution Multi-view Stereo Depth Inference
1912.06378 (2019-12)
Cascade Cost Volume for High-Resolution Multi-View Stereo and Stereo Matching
1912.11744 (2019-12)
Planar Prior Assisted PatchMatch Multi-View Stereo
and 18 more not named here
read_cameras_binary 15 other papers
13 dated earlier, 2 later
2207.09137 (2022-07)
ParticleSfM: Exploiting Dense Point Trajectories for Localizing Moving Cameras in the Wild
2304.09987 (2023-04)
Tetra-NeRF: Representing Neural Radiance Fields Using Tetrahedra
2305.17398 (2023-05)
NeRO: Neural Geometry and BRDF Reconstruction of Reflective Objects from Multiview Images
and 12 more not named here

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

We present a system that allows for accurate, fast, and robust estimation of camera parameters and depth maps from casual monocular videos of dynamic scenes. Most conventional structure from motion and monocular SLAM techniques assume input videos that feature predominantly static scenes with large amounts of parallax. Such methods tend to produce erroneous estimates in the absence of these conditions. Recent neural network-based approaches attempt to overcome these challenges; however, such methods are either computationally expensive or brittle when run on dynamic videos with uncontrolled camera motion or unknown field of view. We demonstrate the surprising effectiveness of a deep visual SLAM framework: with careful modifications to its training and inference schemes, this system can scale to real-world videos of complex dynamic scenes with unconstrained camera paths, including videos with little camera parallax. Extensive experiments on both synthetic and real videos demonstrate that our system is significantly more accurate and robust at camera pose and depth estimation when compared with prior and concurrent work, with faster or comparable running times. See interactive results on our project page: https://mega-sam.github.io/

For agents

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

get_harvested_code_for_paper("2412.04463")
get_code_for_paper("2412.04463")
have("2412.04463")

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

curl https://syntology.ai/api/ran/2412.04463.json

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

[![Syntology run record](https://syntology.ai/api/ran/2412.04463.svg)](https://syntology.ai/paper/2412.04463)

Connect an agent — have() is free.