SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2203.04913 · CVPR · 2022

Leveling Down in Computer Vision: Pareto Inefficiencies in Fair Deep Classifiers

Francesco Locatello, Bernhard Schölkopf, Michael Lohaus, Matthäus Kleindessner, Chris Russell, Dominik Zietlow, Guha Balakrishnan

arXiv · PDF · Open in the Atlas

Code that ran

9 of 12 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; 1 honoured the contract we drafted, 0 violated it and 2 have no contract verdict; 6 failures of our instrument; 3 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/2203.04913.json · badge: /api/ran/2203.04913.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
oxfordinternetinstitute/oxonfair community
Papers with Code archive
9 ran · 3 no recorded run
FunctionStatusWhere it lives
ScorerRequiresContPred Ran
ran on a synthesized input · constructed an object; no result was computed
oxfordinternetinstitute/oxonfair/src/oxonfair/learners/fair_frontier.py
code served (permissive licence) · get_code("a798938602634eac")
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, ScorerRequiresContPred: 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
from typing import Any

torch.manual_seed(0)

class DummyScorer(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 2)
    
    def forward(self, predictions, targets):
        # Simple dummy scorer that takes predictions and targets
        # and returns a scalar score
        return torch.tensor(0.85)

# Create a simple scorer callable
scorer = DummyScorer()

# Create ARGS dict with all parameters of ScorerRequiresContPred.__init__
ARGS = {
    "scorer": scorer,
    "name": "test_scorer"
}
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 ScorerRequiresContPred — 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 ScorerRequiresContPred 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
inv_sigmoid Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
oxfordinternetinstitute/oxonfair/src/oxonfair/learners/fair_frontier.py
code served (permissive licence) · get_code("a0f7783ca53b47b4")
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 (2, 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 numpy as np

# Create a valid input for inv_sigmoid
# The function expects x to be strictly between 0 and 1
# Let's create a small numpy array with values in (0, 1)
x = np.array([[0.1, 0.5, 0.9, 0.3],
              [0.2, 0.7, 0.4, 0.8]])

ARGS = {
    "x": x
}
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 (2, 4) — shape (2, 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
keep_front Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
oxfordinternetinstitute/oxonfair/src/oxonfair/learners/fair_frontier.py
code served (permissive licence) · get_code("77be5e697751a0e2")
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 numpy as np

# Set random seed for reproducibility
np.random.seed(0)

# Create small test data for keep_front function
# The function expects:
# - solutions: 2D numpy array of shape (n_objectives, n_solutions)
# - initial_weights: 2D or 3D numpy array of shape (n_solutions,) or (batch, n_solutions, n_weights) or similar
# - directions: 1D numpy array of shape (n_objectives,) with values +1 or -1
# - additional_constraints: sequence of floats, size = n_objectives - 2

# Create solutions array (objectives x solutions)
# Shape: (4, 6) - 4 objectives, 6 solutions
solutions = np.array([
    [1.0, 2.0, 3.0, 4.0, 5.0, 6.0],
    [2.0, 3.0, 4.0, 5.0, 6.0, 7.0],
    [0.5, 1.0, 1.5, 2.0, 2.5, 3.0],
    [0.1, 0.2, 0.3, 0.4, 0.5, 0.6]
], dtype=np.float64)

# Create initial weights (6 solutions, 3 weight dimensions)
initial_weights = np.array([
    [0.1, 0.2, 0.3],
    [0.15, 0.25, 0.35],
    [0.12, 0.22, 0.32],
    [0.18, 0.28, 0.38],
    [0.14, 0.24, 0.34],
    [0.16, 0.26, 0.36]
], dtype=np.float64).T  # Transpose to (3, 6)

# Create directions vector (maximize first two, minimize last two)
# +1 means maximize, -1 means minimize
directions = np.array([1, 1, -1, -1], dtype=np.float64)

# Create additional constraints (size = n_objectives - 2 = 2)
# These constrain objectives at indices 2 and 3
additional_constraints = [0.5, 0.2]

ARGS = {
    'solutions': solutions,
    'initial_weights': initial_weights,
    'directions': directions,
    'additional_constraints': additional_constraints,
    'force_levelling_up': False
}
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 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
linear_interpolate Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
oxfordinternetinstitute/oxonfair/src/oxonfair/learners/fair_frontier.py
code served (permissive licence) · get_code("564041634dc60656")
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 (2, 3, 103) 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

# Analyze the function:
# - front: np.ndarray, used with .max(1), .min(1), indexing [:, i], [:, i+1]
#   Shape should be (batch, num_points) where num_points > 1
# - weights: np.ndarray, used with [:, :, i], [:, :, i+1], .shape[-1]
#   Shape should be (batch, some_dim, num_points) where num_points > 1
# - gap: float, default 0.01

# Create front array: shape (2, 5) - 2 batch items, 5 frontier points
front = np.array([
    [1.0, 2.0, 3.5, 5.0, 6.5],
    [0.5, 1.5, 3.0, 4.5, 6.0]
], dtype=np.float32)

# Create weights array: shape (2, 3, 5) - 2 batch items, 3 weight dims, 5 points
weights = np.random.randn(2, 3, 5).astype(np.float32)

# gap parameter
gap = 0.01

ARGS = {
    "front": front,
    "weights": weights,
    "gap": gap
}
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 (2, 3, 103) — shape (2, 3, 103) 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
make_finer_grid Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
oxfordinternetinstitute/oxonfair/src/oxonfair/learners/fair_frontier.py
code served (permissive licence) · get_code("9f17ba1fc0db9998")
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 (2, 4, 2917) 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 a small weights array
# Shape: (groups, classes + 1, num_points)
# Following the pattern in make_finer_grid, weights should have shape (groups, classes + 1, num_points)
# where the function processes weights[:, :-1, ell] and weights[:, :-1, ell + 1]

groups = 2
classes = 3
num_points = 5

weights = np.random.randn(groups, classes + 1, num_points).astype(np.float16)

# Create ARGS dict with all parameters of make_finer_grid
ARGS = {
    "weights": weights,
    "refinement_factor": 2,
    "use_linspace": True,
}
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 (2, 4, 2917) — shape (2, 4, 2917) 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
make_grid_arange Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
oxfordinternetinstitute/oxonfair/src/oxonfair/learners/fair_frontier.py
code served (permissive licence) · get_code("f6611e085260a40e")
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 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 numpy as np

# Construct arguments for make_grid_arange
mins = np.array([0.0, 1.0, 2.0, 3.0])
maxs = np.array([10.0, 11.0, 12.0, 13.0])
refinement_factor = np.array([0.5, 0.5, 0.5, 0.5])
add_zero = True

ARGS = {
    "mins": mins,
    "maxs": maxs,
    "refinement_factor": refinement_factor,
    "add_zero": add_zero,
}
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 list, ndim 2, 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
make_grid_between_points Ran
ran on a synthesized input
oxfordinternetinstitute/oxonfair/src/oxonfair/learners/fair_frontier.py
code served (permissive licence) · get_code("a63e6e87527b81a5")
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 (2, 5, 65536) 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

# Construct valid arguments for make_grid_between_points
point_a = np.array([[0.1, 0.2, 0.3, 0.4],
                     [0.15, 0.25, 0.35, 0.45]])
point_b = np.array([[0.5, 0.6, 0.7, 0.8],
                     [0.55, 0.65, 0.75, 0.85]])

refinement_factor = 3

ARGS = {
    "point_a": point_a,
    "point_b": point_b,
    "refinement_factor": refinement_factor,
    "add_zero": False,
    "use_linspace": True,
    "logit_scaling": False
}
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 (2, 5, 65536) — shape (2, 5, 65536) 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
make_grid_linspace Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
oxfordinternetinstitute/oxonfair/src/oxonfair/learners/fair_frontier.py
code served (permissive licence) · get_code("b967ef0154b5c690")
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 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 numpy as np

# Create test arguments for make_grid_linspace
mins = np.array([0.1, 0.2, 0.3, 0.15])
maxs = np.array([0.8, 0.9, 0.7, 0.85])
logit_scaling = False
refinement_factor = 5
add_zero = True

ARGS = {
    "mins": mins,
    "maxs": maxs,
    "logit_scaling": logit_scaling,
    "refinement_factor": refinement_factor,
    "add_zero": add_zero,
}
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 list, ndim 2, 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
sigmoid Ran
ran, honoured the contract we drafted
oxfordinternetinstitute/oxonfair/src/oxonfair/learners/fair_frontier.py
code served (permissive licence) · get_code("d05c71a4b943dde3")
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 (2, 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 numpy as np

# Create a small numpy array to pass to sigmoid
x = np.array([[-2.0, -1.0, 0.0, 1.0],
              [2.0, 0.5, -0.5, 1.5]])

ARGS = {
    "x": x
}
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 (2, 4) — shape (2, 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
build_coarse_to_fine_front Not yet run
unverified (harvested, no recorded run)
oxfordinternetinstitute/oxonfair/src/oxonfair/learners/fair_frontier.py
code served (permissive licence) · get_code("3db1bcc322e1063f")
How it ran

no invocation recorded.

compute_metrics Not yet run
unverified (harvested, no recorded run)
oxfordinternetinstitute/oxonfair/src/oxonfair/learners/fair_frontier.py
code served (permissive licence) · get_code("da52ff2f71de7630")
How it ran

no invocation recorded.

front_from_weights Not yet run
unverified (harvested, no recorded run)
oxfordinternetinstitute/oxonfair/src/oxonfair/learners/fair_frontier.py
code served (permissive licence) · get_code("9b5a3d2ae57a9889")
How it ran

no invocation recorded.

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 12 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 12 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

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

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

Repositories linked to this paper

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

Abstract

Algorithmic fairness is frequently motivated in terms of a trade-off in which overall performance is decreased so as to improve performance on disadvantaged groups where the algorithm would otherwise be less accurate. Contrary to this, we find that applying existing fairness approaches to computer vision improve fairness by degrading the performance of classifiers across all groups (with increased degradation on the best performing groups). Extending the bias-variance decomposition for classification to fairness, we theoretically explain why the majority of fairness methods designed for low capacity models should not be used in settings involving high-capacity models, a scenario common to computer vision. We corroborate this analysis with extensive experimental support that shows that many of the fairness heuristics used in computer vision also degrade performance on the most disadvantaged groups. Building on these insights, we propose an adaptive augmentation strategy that, uniquely, of all methods tested, improves performance for the disadvantaged groups.

For agents

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

get_harvested_code_for_paper("2203.04913")
get_code_for_paper("2203.04913")
have("2203.04913")

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

curl https://syntology.ai/api/ran/2203.04913.json

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

[![Syntology run record](https://syntology.ai/api/ran/2203.04913.svg)](https://syntology.ai/paper/2203.04913)

Connect an agent — have() is free.