SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2404.08351 · ECCV · 2024

OmniSat: Self-Supervised Modality Fusion for Earth Observation

Nicolas Gonthier, Loic Landrieu, Guillaume Astruc, Clement Mallet

arXiv · PDF · Open in the Atlas

Code that ran

14 of 23 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (1 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 1 has no contract verdict; 11 failures of our instrument; 9 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/2404.08351.json · badge: /api/ran/2404.08351.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
gastruc/omnisat canonical (classifier label)
classifier basis: census
2 ran · 0 no recorded run
gastruc/OmniSat community
Papers with Code archive
12 ran · 9 no recorded run
FunctionStatusWhere it lives
METHOD Ran
ran on a synthesized input · constructed an object; no result was computed
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("6f2776fda8eaa86b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, METHOD: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

class METHOD:
    """define iRPE method IDs
    We divide the implementation of CROSS into CROSS_ROWS and CROSS_COLS.

    """
    EUCLIDEAN = 0
    QUANT = 1
    PRODUCT = 3
    CROSS = 4
    CROSS_ROWS = 41
    CROSS_COLS = 42


ARGS = {}
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 METHOD — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a METHOD 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
_rp_2d_cross_cols Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("3c90a55d42ebd29a")
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 2, dtype family int, shape (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
import math

torch.manual_seed(0)

# Create a small 2D difference tensor with shape (L, L, 2)
# where L=3 for a small sequence length
# The last dimension of size 2 represents (row_offset, col_offset)
L = 3
diff = torch.randn(L, L, 2, dtype=torch.float32)

# Set up the parameters for piecewise_index function
alpha = 2.0
beta = 4.0
gamma = 3.0
dtype = torch.long

ARGS = {
    "diff": diff,
    "alpha": alpha,
    "beta": beta,
    "gamma": gamma,
    "dtype": dtype,
}
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 2, dtype family int, shape (3, 3) — shape (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
_rp_2d_cross_rows Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("afa68c96506b7ca4")
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 2, dtype family int, shape (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
import math

torch.manual_seed(0)

# Create a small 2D difference tensor with shape (L, L, 2)
# L=3 for a small sequence length
L = 3
diff = torch.randn(L, L, 2, dtype=torch.float32)

# Define the kwargs that will be passed to piecewise_index
kwargs = {
    'alpha': 2.0,
    'beta': 4.0,
    'gamma': 1.2,
    'dtype': torch.long
}

ARGS = {
    'diff': diff,
    **kwargs
}
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 2, dtype family int, shape (3, 3) — shape (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
_rp_2d_euclidean Ran
ran; our contract draft was wrong, not the code
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("2afb88f76e6694db")
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 2, dtype family int, shape (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
import math

torch.manual_seed(0)

# Create a small 2D difference tensor
# Shape: (L, L, 2) where L=3, representing row and column offsets
L = 3
diff = torch.randn(L, L, 2)

# Define the kwargs for piecewise_index
alpha = 2.0
beta = 4.0
gamma = 2.0
dtype = torch.long

ARGS = {
    'diff': diff,
    'alpha': alpha,
    'beta': beta,
    'gamma': gamma,
    'dtype': dtype
}
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 2, dtype family int, shape (3, 3) — shape (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
_rp_2d_product Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("5e64867591fc57ca")
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 2, dtype family int, shape (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
import math

torch.manual_seed(0)

# Create a small 2D difference tensor of shape (L, L, 2)
# L=3 for a small sequence length
L = 3
diff = torch.randn(L, L, 2, dtype=torch.float32)

# Define the kwargs parameters needed by _rp_2d_product
# These are passed to piecewise_index
ARGS = {
    'diff': diff,
    'alpha': 1.0,
    'beta': 2.0,
    'gamma': 1.5,
    'dtype': torch.long
}
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 2, dtype family int, shape (3, 3) — shape (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
_rp_2d_quant Ran
ran; our contract draft was wrong, not the code
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("ea730722397f9780")
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 2, dtype family int, shape (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
import math

torch.manual_seed(0)

# Create a small 2D difference tensor with shape (L, L, 2)
# L = 3 for a small sequence length
L = 3
diff = torch.randn(L, L, 2, dtype=torch.float32)

# Define the kwargs that will be passed to piecewise_index
kwargs = {
    'alpha': 2.0,
    'beta': 4.0,
    'gamma': 8.0,
    'dtype': torch.long
}

ARGS = {
    'diff': diff,
    **kwargs
}
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 2, dtype family int, shape (3, 3) — shape (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
get_absolute_positions Ran
ran, honoured the contract we drafted
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("ba23bbb042430585")
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 (3, 4, 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 torch

torch.manual_seed(0)

ARGS = {
    "height": 3,
    "width": 4,
    "dtype": torch.float32,
    "device": torch.device("cpu"),
}
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 (3, 4, 2) — shape (3, 4, 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
get_bucket_ids_2d_without_skip Ran
ran; our contract draft was wrong, not the code
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("ab89b48444865c77")
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 math

torch.manual_seed(0)

# Define the METHOD class
class METHOD:
    EUCLIDEAN = 0
    QUANT = 1
    PRODUCT = 3
    CROSS = 4
    CROSS_ROWS = 41
    CROSS_COLS = 42

# Define piecewise_index function (stub implementation)
@torch.no_grad()
def piecewise_index(relative_position, alpha, beta, gamma, dtype):
    """Stub implementation of piecewise_index"""
    beta_int = int(beta)
    # Simple bucketing: clamp and offset
    bucketed = torch.clamp(relative_position, -beta_int, beta_int).long() + beta_int
    return bucketed

# Define helper functions used by get_bucket_ids_2d_without_skip
def get_absolute_positions(height, width, dtype, device):
    """Get absolute positions"""
    rows = torch.arange(height, dtype=dtype, device=device).view(height, 1).repeat(1, width)
    cols = torch.arange(width, dtype=dtype, device=device).view(1, width).repeat(height, 1)
    return torch.stack([rows, cols], 2)

@torch.no_grad()
def _rp_2d_cross_cols(diff, **kwargs):
    """2D RPE with Cross for columns."""
    dis = diff[:, :, 1]
    return piecewise_index(dis, **kwargs)

@torch.no_grad()
def _rp_2d_cross_rows(diff, **kwargs):
    """2D RPE with Cross for rows."""
    dis = diff[:, :, 0]
    return piecewise_index(dis, **kwargs)

@torch.no_grad()
def _rp_2d_euclidean(diff, **kwargs):
    """2D RPE with Euclidean method."""
    dis = diff.square().sum(2).float().sqrt().round()
    return piecewise_index(dis, **kwargs)

@torch.no_grad()
def _rp_2d_product(diff, **kwargs):
    """2D RPE with Product method."""
    beta_int = int(kwargs['beta'])
    S = 2 * beta_int + 1
    r = piecewise_index(diff[:, :, 0], **kwargs) + beta_int
    c = piecewise_index(diff[:, :, 1], **kwargs) + beta_int
    pid = r * S + c
    return pid

@torch.no_grad()
def _rp_2d_quant(diff, **kwargs):
    """2D RPE with Quantization method."""
    dis = diff.square().sum(2)
    return piecewise_index(dis, **kwargs)

# Now define get_bucket_ids_2d_without_skip (minimal stub)
@torch.no_grad()
def get_bucket_ids_2d_without_skip(height, width, method, alpha, beta, gamma, dtype=torch.long, device=torch.device('cpu')):
    """Get 2D bucket IDs without skip token."""
    # Get absolute positions
    positions = get_absolute_positions(height, width, dtype=torch.float32, device=device)
    
    # Compute relative differences
    pos_row = positions[:, :, 0:1]
    pos_col = positions[:, :, 1:2]
    
    diff_row = pos_row - pos_row.transpose(0, 1)
    diff_col = pos_col - pos_col.transpose(0, 1)
    diff = torch.cat([diff_row, diff_col], dim=2)
    
    # Apply appropriate method
    kwargs = {'alpha': alpha, 'beta': beta, 'gamma': gamma, 'dtype': dtype}
    
    if method == METHOD.EUCLIDEAN:
        bucket_ids = _rp_2d_euclidean(diff, **kwargs)
    elif method == METHOD.QUANT:
        bucket_ids = _rp_2d_quant(diff, **kwargs)
    elif method == METHOD.PRODUCT:
        bucket_ids = _rp_2d_product(diff, **kwargs)
    elif method == METHOD.CROSS_ROWS:
        bucket_ids = _rp_2d_cross_rows(diff, **kwargs)
    elif method == METHOD.CROSS_COLS:
        bucket_ids = _rp_2d_cross_cols(diff, **kwargs)
    else:
        raise ValueError(f"Unknown method: {method}")
    
    return bucket_ids

# Create arguments
ARGS = {
    'height': 4,
    'width': 3,
    'method': METHOD.EUCLIDEAN,
    'alpha': 1.0,
    'beta': 2.0,
    'gamma': 1.0,
    'dtype': torch.long,
    'device': torch.device('cpu')
}
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
get_num_buckets Ran
ran, honoured the contract we drafted
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("0deaebd0135ec340")
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 torch
import numpy as np

class METHOD:
    """define iRPE method IDs
    We divide the implementation of CROSS into CROSS_ROWS and CROSS_COLS.

    """
    EUCLIDEAN = 0
    QUANT = 1
    PRODUCT = 3
    CROSS = 4
    CROSS_ROWS = 41
    CROSS_COLS = 42


ARGS = {
    "method": METHOD.PRODUCT,
    "alpha": 0.5,
    "beta": 3.0,
    "gamma": 2.0,
}
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 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_pylogger Ran
ran; our contract draft was wrong, not the code
gastruc/omnisat/src/utils/pylogger.py
code served (permissive licence) · get_code("b0cd1557c2946d8d")
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 Logger. 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
from lightning.pytorch.utilities import rank_zero_only

# The function get_pylogger takes an optional name parameter
# It uses the name to get a logger via logging.getLogger(name)
# The function doesn't actually call the logger methods, it just wraps them
# So we just need to pass a valid string name

ARGS = {
    "name": "test_logger"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
lightning — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Logger — a Logger output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Logger 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
get_rpe_config Ran
ran; our contract draft was wrong, not the code
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("8abf0de547843739")
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 EasyDict. 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 easydict import EasyDict as edict

torch.manual_seed(0)

class METHOD:
    """define iRPE method IDs
    We divide the implementation of CROSS into CROSS_ROWS and CROSS_COLS.

    """
    EUCLIDEAN = 0
    QUANT = 1
    PRODUCT = 3
    CROSS = 4
    CROSS_ROWS = 41
    CROSS_COLS = 42

def get_num_buckets(method, alpha, beta, gamma):
    beta_int = int(beta)
    if method == METHOD.PRODUCT:
        num_buckets = (2 * beta_int + 1) ** 2
    else:
        num_buckets = 2 * beta_int + 1
    return num_buckets

def get_single_rpe_config(ratio=1.9, method=METHOD.PRODUCT, mode='contextual', shared_head=True, skip=0):
    config = edict()
    config.shared_head = shared_head
    config.mode = mode
    config.method = method
    config.alpha = 1 * ratio
    config.beta = 2 * ratio
    config.gamma = 8 * ratio
    config.num_buckets = get_num_buckets(method, config.alpha, config.beta, config.gamma)
    if skip > 0:
        config.num_buckets += 1
    return config

def get_rpe_config(ratio=1.9, method=METHOD.PRODUCT, mode='contextual', shared_head=True, skip=0, rpe_on='k'):
    if isinstance(method, str):
        method_mapping = dict(euc=METHOD.EUCLIDEAN, quant=METHOD.QUANT, cross=METHOD.CROSS, product=METHOD.PRODUCT)
        method = method_mapping[method.lower()]
    if mode == 'ctx':
        mode = 'contextual'
    config = edict()
    kwargs = dict(ratio=ratio, method=method, mode=mode, shared_head=shared_head, skip=skip)
    config.rpe_q = get_single_rpe_config(**kwargs) if 'q' in rpe_on else None
    config.rpe_k = get_single_rpe_config(**kwargs) if 'k' in rpe_on else None
    config.rpe_v = get_single_rpe_config(**kwargs) if 'v' in rpe_on else None
    return config

ARGS = {
    'ratio': 1.9,
    'method': METHOD.PRODUCT,
    'mode': 'contextual',
    'shared_head': True,
    'skip': 0,
    'rpe_on': 'k'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
easydict, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EasyDict — a EasyDict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a EasyDict 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
get_single_rpe_config Ran
ran; our contract draft was wrong, not the code
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("c637a66e6e3977d2")
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 EasyDict. 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 easydict import EasyDict as edict

torch.manual_seed(0)

class METHOD:
    """define iRPE method IDs
    We divide the implementation of CROSS into CROSS_ROWS and CROSS_COLS.

    """
    EUCLIDEAN = 0
    QUANT = 1
    PRODUCT = 3
    CROSS = 4
    CROSS_ROWS = 41
    CROSS_COLS = 42

def get_num_buckets(method, alpha, beta, gamma):
    """ Get number of buckets storing relative position encoding.
    The buckets does not contain `skip` token.

    Parameters
    ----------
    method: METHOD
        The method ID of image relative position encoding.
    alpha, beta, gamma: float
        The coefficients of piecewise index function.

    Returns
    -------
    num_buckets: int
        The number of buckets storing relative position encoding.
    """
    beta_int = int(beta)
    if method == METHOD.PRODUCT:
        num_buckets = (2 * beta_int + 1) ** 2
    else:
        num_buckets = 2 * beta_int + 1
    return num_buckets

def get_single_rpe_config(ratio=1.9, method=METHOD.PRODUCT, mode='contextual', shared_head=True, skip=0):
    """Get the config of single relative position encoding

    Parameters
    ----------
    ratio: float
        The ratio to control the number of buckets.
    method: METHOD
        The method ID of image relative position encoding.
        The `METHOD` class is defined in `irpe.py`.
    mode: str or None
        The mode of image relative position encoding.
        Choices: [None, 'bias', 'contextual']
    shared_head: bool
        Whether to share weight among different heads.
    skip: int
        The number of skip token before spatial tokens.
        When skip is 0, no classification token.
        When skip is 1, there is a classification token before spatial tokens.
        When skip > 1, there are `skip` extra tokens before spatial tokens.

    Returns
    -------
    config: RPEConfig
        The config of single relative position encoding.
    """
    config = edict()
    config.shared_head = shared_head
    config.mode = mode
    config.method = method
    config.alpha = 1 * ratio
    config.beta = 2 * ratio
    config.gamma = 8 * ratio
    config.num_buckets = get_num_buckets(method, config.alpha, config.beta, config.gamma)
    if skip > 0:
        config.num_buckets += 1
    return config

ARGS = {
    'ratio': 1.9,
    'method': METHOD.PRODUCT,
    'mode': 'contextual',
    'shared_head': True,
    'skip': 0
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
easydict, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EasyDict — a EasyDict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a EasyDict 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
piecewise_index Ran
ran; our contract draft was wrong, not the code
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("c69948747b50336f")
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 2, dtype family float, shape (4, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import math

torch.manual_seed(0)

# Create a small relative position tensor (L x L)
L = 4
relative_position = torch.randn(L, L)

# Define scalar coefficients
alpha = 2.0
beta = 4.0
gamma = 8.0

# Define output dtype
dtype = torch.float32

ARGS = {
    "relative_position": relative_position,
    "alpha": alpha,
    "beta": beta,
    "gamma": gamma,
    "dtype": dtype,
}
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 2, dtype family float, shape (4, 4) — shape (4, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
replace_nans_with_mean Ran
ran; our contract draft was wrong, not the code
gastruc/omnisat/src/utils/preprocess_tsaits.py
code served (permissive licence) · get_code("faf9bf7e5bfa997c")
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, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create a batch of images with some NaN values
# Shape: (batch_size, channels, height, width) = (2, 3, 4, 5)
batch_of_images = torch.randn(2, 3, 4, 5)

# Introduce some NaN values
batch_of_images[0, 1, 2, 3] = float('nan')
batch_of_images[1, 0, 0, 1] = float('nan')

ARGS = {
    "batch_of_images": batch_of_images
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 3, 4, 5) — shape (2, 3, 4, 5) 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
CrossRPEAttention Not yet run
unverified (harvested, no recorded run)
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("0a1f41cf0a6f4305")
How it ran

no invocation recorded.

CrossRPEBlock Not yet run
unverified (harvested, no recorded run)
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("5293629e3ba81145")
How it ran

no invocation recorded.

OmniModule Not yet run
unverified (harvested, no recorded run)
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("817465c178ae5eb3")
How it ran

no invocation recorded.

RPEAttention Not yet run
unverified (harvested, no recorded run)
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("ad51e57d0f69b066")
How it ran

no invocation recorded.

RPEBlock Not yet run
unverified (harvested, no recorded run)
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("47ddca6650a8f8cf")
How it ran

no invocation recorded.

build_rpe Not yet run
unverified (harvested, no recorded run)
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("443aa695f6cd8e56")
How it ran

no invocation recorded.

get_bucket_ids_2d Not yet run
unverified (harvested, no recorded run)
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("aa20f508b856826b")
How it ran

no invocation recorded.

iRPE Not yet run
unverified (harvested, no recorded run)
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("58768324e3554ec0")
How it ran

no invocation recorded.

iRPE_Cross Not yet run
unverified (harvested, no recorded run)
gastruc/OmniSat/src/models/networks/encoder/Omni.py
code served (permissive licence) · get_code("f7f7cb2bdbfa66d9")
How it ran

no invocation recorded.

Also present in other papers' code

Of 23 samples lifted from this paper's linked repositories, 23 carry a function identity and 0 do not. Of the 23 distinct functions with an identity, 16 are also present in other papers whose code we hold (16 with at least one paper dated earlier, 1 the same month, 1 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
get_pylogger 14 other papers
8 dated earlier, 1 the same month, 5 later · 1 only as an assembly brick
2209.03143 (2022-09)
AudioLM: a Language Modeling Approach to Audio Generation
2305.14930 (2023-05)
In-Context Impersonation Reveals Large Language Models' Strengths and Biases
2305.20091 (2023-05)
Humans in 4D: Reconstructing and Tracking Humans with Transformers
and 11 more not named here
METHOD
brick of this paper's assembly: helper
2 other papers
2 dated earlier · 2 only as an assembly brick
2107.14222 (2021-07)
Rethinking and Improving Relative Position Encoding for Vision Transformer
2301.08243 (2023-01)
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
_rp_2d_cross_cols
brick of this paper's assembly: helper
2 other papers
2 dated earlier · 1 only as an assembly brick
2107.14222 (2021-07)
Rethinking and Improving Relative Position Encoding for Vision Transformer
2301.08243 (2023-01)
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
_rp_2d_cross_rows
brick of this paper's assembly: helper
2 other papers
2 dated earlier · 1 only as an assembly brick
2107.14222 (2021-07)
Rethinking and Improving Relative Position Encoding for Vision Transformer
2301.08243 (2023-01)
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
_rp_2d_euclidean
brick of this paper's assembly: helper
2 other papers
2 dated earlier · 1 only as an assembly brick
2107.14222 (2021-07)
Rethinking and Improving Relative Position Encoding for Vision Transformer
2301.08243 (2023-01)
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
_rp_2d_product
brick of this paper's assembly: helper
2 other papers
2 dated earlier · 1 only as an assembly brick
2107.14222 (2021-07)
Rethinking and Improving Relative Position Encoding for Vision Transformer
2301.08243 (2023-01)
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
_rp_2d_quant
brick of this paper's assembly: helper
2 other papers
2 dated earlier · 1 only as an assembly brick
2107.14222 (2021-07)
Rethinking and Improving Relative Position Encoding for Vision Transformer
2301.08243 (2023-01)
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
get_absolute_positions
brick of this paper's assembly: helper
2 other papers
2 dated earlier · 2 only as an assembly brick
2107.14222 (2021-07)
Rethinking and Improving Relative Position Encoding for Vision Transformer
2301.08243 (2023-01)
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
get_bucket_ids_2d_without_skip
brick of this paper's assembly: helper
2 other papers
2 dated earlier · 1 only as an assembly brick
2107.14222 (2021-07)
Rethinking and Improving Relative Position Encoding for Vision Transformer
2301.08243 (2023-01)
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
get_num_buckets
brick of this paper's assembly: helper
2 other papers
2 dated earlier · 2 only as an assembly brick
2107.14222 (2021-07)
Rethinking and Improving Relative Position Encoding for Vision Transformer
2301.08243 (2023-01)
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
iRPE
brick of this paper's assembly: helper
2 other papers
2 dated earlier
2107.14222 (2021-07)
Rethinking and Improving Relative Position Encoding for Vision Transformer
2301.08243 (2023-01)
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
piecewise_index
brick of this paper's assembly: helper
2 other papers
2 dated earlier · 1 only as an assembly brick
2107.14222 (2021-07)
Rethinking and Improving Relative Position Encoding for Vision Transformer
2301.08243 (2023-01)
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
build_rpe
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2301.08243 (2023-01)
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
get_bucket_ids_2d
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2107.14222 (2021-07)
Rethinking and Improving Relative Position Encoding for Vision Transformer
get_rpe_config
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2301.08243 (2023-01)
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
iRPE_Cross
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2301.08243 (2023-01)
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture

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

The diversity and complementarity of sensors available for Earth Observations (EO) calls for developing bespoke self-supervised multimodal learning approaches. However, current multimodal EO datasets and models typically focus on a single data type, either mono-date images or time series, which limits their impact. To address this issue, we introduce OmniSat, a novel architecture able to merge diverse EO modalities into expressive features without labels by exploiting their alignment. To demonstrate the advantages of our approach, we create two new multimodal datasets by augmenting existing ones with new modalities. As demonstrated for three downstream tasks-forestry, land cover classification, and crop mapping-OmniSat can learn rich representations without supervision, leading to state-of-the-art performances in semi-and fully supervised settings. Furthermore, our multimodal pretraining scheme improves performance even when only one modality is available for inference. The code and dataset are available at https://github.com/gastruc/OmniSat.

For agents

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

get_harvested_code_for_paper("2404.08351")
get_code_for_paper("2404.08351")
have("2404.08351")

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

curl https://syntology.ai/api/ran/2404.08351.json

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

[![Syntology run record](https://syntology.ai/api/ran/2404.08351.svg)](https://syntology.ai/paper/2404.08351)

Connect an agent — have() is free.