SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2312.03533 · NeurIPS · 2023

Low-shot Object Learning with Mutual Exclusivity Bias

Stefan Stojanov, Anh Thai, Zixuan Huang, James Rehg, Ahmad Humayun, Bikram Boote

arXiv · PDF · Open in the Atlas

Code that ran

8 of 11 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; 1 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 7 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/2312.03533.json · badge: /api/ran/2312.03533.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
rehg-lab/LSME canonical (classifier label)
classifier basis: census
8 ran · 3 no recorded run
FunctionStatusWhere it lives
check_rendering Ran
ran; our contract draft was wrong, not the code
rehg-lab/LSME/data_utils/check_rendering.py
code served (permissive licence) · get_code("6999c939ef18b891")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import tempfile
from PIL import Image

# Create a temporary directory with some test image files
tmpdir = tempfile.mkdtemp()

# Create a valid RGB image and save it
img_valid = Image.new('RGB', (10, 10), color='red')
valid_img_path = os.path.join(tmpdir, 'valid_image.png')
img_valid.save(valid_img_path)

# Create another valid image
img_valid2 = Image.new('RGB', (20, 20), color='blue')
valid_img_path2 = os.path.join(tmpdir, 'valid_image2.jpg')
img_valid2.save(valid_img_path2)

# Create a file that is not a valid image (text file)
invalid_file_path = os.path.join(tmpdir, 'not_an_image.txt')
with open(invalid_file_path, 'w') as f:
    f.write('This is not an image')

ARGS = {
    'spath': tmpdir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
compute_similarity_global Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
rehg-lab/LSME/baselines/models/utils.py
code served (permissive licence) · get_code("65ace2509cfa0e42")
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. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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)

# The function expects:
# - q: 2D or 3D tensor
# - k: 2D tensor (used in transpose)
# - k_neg: not actually used in the function body, but is a parameter

# For the 2D case (len(q.shape) == 2):
# q.shape[0] is batch size, q.shape[-1] is feature dim
# The code reshapes q into (-1, 3, q.shape[-1]), so q.shape[0] must be divisible by 3
# k should be same shape as q for mm(q, k.T) to work

# Let's create tensors for the 2D case
batch_size = 6  # divisible by 3
feature_dim = 4

q = torch.randn(batch_size, feature_dim)
k = torch.randn(batch_size, feature_dim)
k_neg = torch.randn(batch_size, feature_dim)  # not used but required as parameter

ARGS = {
    "q": q,
    "k": k,
    "k_neg": k_neg,
}
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 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
compute_similarity_global_single Ran
ran; our contract draft was wrong, not the code
rehg-lab/LSME/baselines/models/utils.py
code served (permissive licence) · get_code("67a33aff0eaa9121")
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 torch

torch.manual_seed(0)

# The function expects:
# - q: a tensor that will be unsqueezed and used in bmm operations
# - k: a tensor that will be unsqueezed and used in bmm operations
# - k_neg: optional tensor (can be None) that will be unsqueezed and used in bmm operations

# Looking at the function:
# - q.unsqueeze(1) creates shape (batch, 1, dim)
# - k.unsqueeze(2) creates shape (batch, dim, 1)
# - bmm result is (batch, 1, 1)
# - torch.roll(torch.arange(len(q)), 1) uses len(q), so q must have a length (first dimension is batch)
# - q[perm] indexes q, so it must support indexing

# All tensors should have the same batch size and embedding dimension
batch_size = 2
embedding_dim = 4

q = torch.randn(batch_size, embedding_dim)
k = torch.randn(batch_size, embedding_dim)
k_neg = torch.randn(batch_size, embedding_dim)

ARGS = {
    "q": q,
    "k": k,
    "k_neg": k_neg,
}
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 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
get_padded_bbx Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
rehg-lab/LSME/data_utils/data_utils.py
code served (permissive licence) · get_code("6d63a4a5eb054e60")
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 numpy as np

# Create a segmentation mask with some non-zero values
# Shape: (height=20, width=15) - distinct dimensions for probing
seg = np.zeros((20, 15), dtype=np.float32)

# Add some non-zero values in the middle to create a bounding box
seg[5:15, 3:12] = 1.0

# Padding value
pad = 10

ARGS = {
    "seg": seg,
    "pad": pad,
}
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, 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
get_paths Ran
ran; our contract draft was wrong, not the code
rehg-lab/LSME/data_utils/data_utils.py
code served (permissive licence) · get_code("32f1ebc2d1f583c0")
How it ran

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

import os
import tempfile
import json

# Create a temporary directory structure to satisfy the function's assertions
temp_dir = tempfile.mkdtemp()
DATA_ROOT = temp_dir
scene = "test_scene"
instance = "test_instance"
frame = 0
ext = "png"

# Create the necessary directory structure
os.makedirs(os.path.join(DATA_ROOT, scene, 'depth_NPZ'), exist_ok=True)
os.makedirs(os.path.join(DATA_ROOT, scene, 'RGB'), exist_ok=True)
os.makedirs(os.path.join(DATA_ROOT, scene, 'segmentations', instance), exist_ok=True)

# Create the required files
depth_file = os.path.join(DATA_ROOT, scene, 'depth_NPZ', '{:04d}.npz'.format(frame))
with open(depth_file, 'w') as f:
    f.write('')

img_file = os.path.join(DATA_ROOT, scene, 'RGB', '{:04d}.png'.format(frame))
with open(img_file, 'w') as f:
    f.write('')

seg_file = os.path.join(DATA_ROOT, scene, 'segmentations', instance, '{}_{:04d}.png'.format(instance, frame))
with open(seg_file, 'w') as f:
    f.write('')

meta_file = os.path.join(DATA_ROOT, scene, 'metadata_updated.json')
with open(meta_file, 'w') as f:
    json.dump({}, f)

ARGS = {
    'DATA_ROOT': DATA_ROOT,
    'scene': scene,
    'instance': instance,
    'frame': frame,
    'ext': ext
}
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 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
get_pixel_grid Ran
ran, honoured the contract we drafted
rehg-lab/LSME/data_utils/data_utils.py
code served (permissive licence) · get_code("ca375db15490350b")
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 int, shape (20, 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

# Analyze the function:
# - H: height (int), used in np.arange(H)
# - W: width (int), used in np.arange(W)
# Both are simple integers used to create coordinate grids

ARGS = {
    "H": 4,
    "W": 5,
}
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 int, shape (20, 3) — shape (20, 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_statistics Ran
ran; our contract draft was wrong, not the code
rehg-lab/LSME/data_generation/check_rendered_data.py
code served (permissive licence) · get_code("d5e33b6626b083a9")
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 os
import tempfile
import shutil

# Create temporary directories with the expected structure
temp_dir1 = tempfile.mkdtemp()
temp_dir2 = tempfile.mkdtemp()

# Create segmentations subdirectories
seg_dir1 = os.path.join(temp_dir1, 'segmentations')
seg_dir2 = os.path.join(temp_dir2, 'segmentations')
os.makedirs(seg_dir1, exist_ok=True)
os.makedirs(seg_dir2, exist_ok=True)

# Create some dummy segmentation files
for i in range(3):
    open(os.path.join(seg_dir1, f'seg_{i}.npy'), 'w').close()

for i in range(2):
    open(os.path.join(seg_dir2, f'seg_{i}.npy'), 'w').close()

# Create a floor_object file that should be ignored
open(os.path.join(seg_dir1, 'floor_object'), 'w').close()

ARGS = {
    'arg': [temp_dir1, temp_dir2]
}
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
job Ran
ran; our contract draft was wrong, not the code
rehg-lab/LSME/data_utils/check_rendering.py
code served (permissive licence) · get_code("4f9430a300acd81e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import tempfile
from PIL import Image

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

# Create a few valid RGB image files
img1 = Image.new('RGB', (10, 10), color='red')
img1.save(os.path.join(temp_dir, 'image1.png'))

img2 = Image.new('RGB', (20, 20), color='blue')
img2.save(os.path.join(temp_dir, 'image2.jpg'))

# Create one file that will cause an error (corrupted/invalid image)
with open(os.path.join(temp_dir, 'corrupted.png'), 'wb') as f:
    f.write(b'not a valid image file')

ARGS = {
    'arg': temp_dir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
batch_shuffle_ddp Not yet run
unverified (harvested, no recorded run)
rehg-lab/LSME/baselines/nets/moco_func_utils.py
code served (permissive licence) · get_code("35c74641cae2d081")
How it ran

no invocation recorded.

batch_unshuffle_ddp Not yet run
unverified (harvested, no recorded run)
rehg-lab/LSME/baselines/nets/moco_func_utils.py
code served (permissive licence) · get_code("7573bfb6ba2070ce")
How it ran

no invocation recorded.

concat_all_gather Not yet run
unverified (harvested, no recorded run)
rehg-lab/LSME/baselines/nets/moco_func_utils.py
code served (permissive licence) · get_code("73cecca9f3575f09")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 2 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 1 the same month, 1 later) and 9 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
concat_all_gather 138 other papers
96 dated earlier, 4 the same month, 37 later, 1 same year, order not determinable · 33 only as an assembly brick
1710.09412 (2017-10)
mixup: BEYOND EMPIRICAL RISK MINIMIZATION
1905.04899 (2019-05)
CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features
1905.12794 (2019-05)
Fashion IQ: A New Dataset Towards Retrieving Images by Natural Language Feedback
and 135 more not named here
batch_shuffle_ddp 2 other papers
2 dated earlier
2103.16607 (2021-03)
Seasonal Contrast: Unsupervised Pre-Training from Uncurated Remote Sensing Data
2211.15059 (2022-11)
Learning Dense Object Descriptors from Multiple Views for Low-shot Category Generalization

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

This paper introduces Low-shot Object Learning with Mutual Exclusivity Bias (LSME), the first computational framing of mutual exclusivity bias, a phenomenon commonly observed in infants during word learning. We provide a novel dataset, comprehensive baselines, and a state-of-the-art method to enable the ML community to tackle this challenging learning task. The goal of LSME is to analyze an RGB image of a scene containing multiple objects and correctly associate a previously-unknown object instance with a provided category label. This association is then used to perform low-shot learning to test category generalization. We provide a data generation pipeline for the LSME problem and conduct a thorough analysis of the factors that contribute to its difficulty. Additionally, we evaluate the performance of multiple baselines, including state-of-the-art foundation models. Finally, we present a baseline approach that outperforms state-of-the-art models in terms of low-shot accuracy. Code and data are available at https://github.com/rehglab/LSME. lion car sandwich lion sandwich car * Equal contribution. 37th Conference on Neural Information Processing Systems (NeurIPS 2023) Track on Datasets and Benchmarks.

For agents

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

get_harvested_code_for_paper("2312.03533")
get_code_for_paper("2312.03533")
have("2312.03533")

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

curl https://syntology.ai/api/ran/2312.03533.json

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

[![Syntology run record](https://syntology.ai/api/ran/2312.03533.svg)](https://syntology.ai/paper/2312.03533)

Connect an agent — have() is free.