SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1805.07932 · 2018

Bilinear Attention Networks

arXiv · PDF · Open in the Atlas

Code that ran

13 of 13 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, 3 violated it and 8 have no contract verdict; 1 failure of our instrument; 0 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/1805.07932.json · badge: /api/ran/1805.07932.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
jnhwkim/ban-vqa canonical (classifier label)
classifier basis: census
7 ran · 0 no recorded run
Cyanogenoid/vqa-counting community (classifier label)
classifier basis not recorded on the link
6 ran · 0 no recorded run
FunctionStatusWhere it lives
answer_filter Ran
ran, violated the contract we drafted
jnhwkim/ban-vqa/dataset.py
pointer only (licence: MIT); the stored copy get_code serves, from KaveeshaSIlva/PathVQA, is pointer only (licence: NONE) · get_code("cf2efa825f4afaf0")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from KaveeshaSIlva/PathVQA, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

apply_attention Ran
ran on a synthesized input
Cyanogenoid/vqa-counting/vqa-v2/model.py
code served (permissive licence) · get_code("26e02979502211f1")
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 (2, 20) 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 torch.nn.functional as F

torch.manual_seed(0)

# Create input tensor: (batch_size=2, channels=4, spatial_dim=3)
# Shape: [n, c, spatial_size] where spatial_size will be flattened
input_tensor = torch.randn(2, 4, 3)

# Create attention tensor: (batch_size=2, glimpses=5, spatial_dim=3)
# Shape: [n, glimpses, spatial_size] where spatial_size matches input
attention_tensor = torch.randn(2, 5, 3)

ARGS = {
    "input": input_tensor,
    "attention": attention_tensor,
}
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 (2, 20) — shape (2, 20) 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
batch_accuracy Ran
ran on a synthesized input
Cyanogenoid/vqa-counting/vqa-v2/utils.py
code served (permissive licence) · get_code("2c844ba6b5d8ece2")
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 (2, 1) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create small tensors for batch_accuracy
# predicted: logits with shape (batch_size, num_classes)
# true: binary labels for each class with shape (batch_size, num_classes)
batch_size = 2
num_classes = 4

predicted = torch.randn(batch_size, num_classes)
true = torch.randint(0, 2, (batch_size, num_classes))

ARGS = {
    "predicted": predicted,
    "true": true,
}
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 (2, 1) — shape (2, 1) 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
collate_fn Ran
ran on a synthesized input
Cyanogenoid/vqa-counting/vqa-v2/data.py
code served (permissive licence) · get_code("28e89c10480659aa")
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. 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.utils.data as data

torch.manual_seed(0)

# Create a small batch of data
# Each item in the batch is a tuple where the last element is used for sorting
# We'll create tuples of (tensor, label) where label is numeric for sorting
batch = [
    (torch.randn(4, 3), 5),
    (torch.randn(4, 3), 2),
    (torch.randn(4, 3), 8),
    (torch.randn(4, 3), 1),
]

ARGS = {
    "batch": batch
}
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 list — a list 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
compute_recall_with_logits Ran
ran on a synthesized input
jnhwkim/ban-vqa/train_flickr.py
code served (permissive licence) · get_code("5a8370941ac95d70")
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. 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)

# Create small tensors for testing
batch_size = 2
num_classes = 4

# logits: shape (batch_size, num_classes) - raw scores for ranking
logits = torch.randn(batch_size, num_classes)

# labels: shape (batch_size, num_classes) - multi-hot encoded labels
labels = torch.zeros(batch_size, num_classes)
labels[0, 0] = 1  # first sample has class 0 as positive
labels[0, 2] = 1  # first sample has class 2 as positive
labels[1, 1] = 1  # second sample has class 1 as positive

ARGS = {
    "logits": logits,
    "labels": labels,
}
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 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
evaluate Ran
ran on a synthesized input
jnhwkim/ban-vqa/train_flickr.py
code served (permissive licence) · get_code("728c76c97b1bb97a")
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. 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 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)

# Build a minimal model that satisfies the function's usage
class DummyModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 5)
    
    def forward(self, v, b, p, e, unused):
        # Returns (something, logits)
        # logits shape should be (batch, seq_len, n_obj) based on usage
        batch_size = v.size(0)
        logits = self.linear(v)  # (batch, 5)
        logits = logits.unsqueeze(1).expand(batch_size, 3, 5)  # (batch, 3, 5)
        return None, logits

model = DummyModel()

# Build a minimal dataloader
# Each batch has: (v, b, p, e, n, a, idx, types)
batch_size = 2
seq_len = 3
n_obj = 5
feature_dim = 4

# v: visual features (batch, feature_dim)
v = torch.randn(batch_size, feature_dim)

# b: bounding boxes (batch, ...)
b = torch.randn(batch_size, 4)

# p: positions (batch, ...)
p = torch.randn(batch_size, 2)

# e: embeddings (batch, ...)
e = torch.randn(batch_size, feature_dim)

# n: number of objects per sample (batch, 1) - contains counts for each sample
n = torch.tensor([[seq_len], [seq_len]], dtype=torch.long)

# a: answer/label tensor (batch, seq_len, n_obj)
a = torch.zeros(batch_size, seq_len, n_obj)
a[0, 0, 1] = 1  # Set some non-zero values
a[1, 1, 2] = 1

# idx: indices
idx = torch.tensor([0, 1])

# types: types
types = torch.tensor([0, 1])

# Create dataloader as list of batches
dataloader = [
    (v, b, p, e, n, a, idx, types)
]

ARGS = {
    'model': model,
    'dataloader': dataloader
}
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
extract_vocab Ran
ran on a synthesized input
Cyanogenoid/vqa-counting/vqa-v2/preprocess-vocab.py
code served (permissive licence) · get_code("8e41785566e403d2")
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.

from collections import Counter
import itertools

# Construct test data for extract_vocab
# The function expects:
# - iterable: an iterable of lists of tokens
# - top_k: optional int for filtering to top k most common tokens
# - start: int for starting index in vocab dict

# Create sample token lists (lists of strings representing tokens)
token_lists = [
    ['apple', 'banana', 'apple', 'cherry'],
    ['banana', 'apple', 'date'],
    ['cherry', 'date', 'apple'],
    ['apple', 'apple', 'banana']
]

ARGS = {
    'iterable': token_lists,
    'top_k': None,
    'start': 0
}
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
is_howmany Ran
ran, violated the contract we drafted
jnhwkim/ban-vqa/dataset.py
pointer only (licence: MIT); the stored copy get_code serves, from KaveeshaSIlva/PathVQA, is pointer only (licence: NONE) · get_code("ac9f54e37ab206e4")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from KaveeshaSIlva/PathVQA, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

load_folder Ran
ran; our contract draft was wrong, not the code
jnhwkim/ban-vqa/utils.py
code served (permissive licence) · get_code("8c1252eacab71602")
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
import shutil

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

# Create some test files with different suffixes
test_files = [
    "image_001.jpg",
    "image_002.jpg",
    "image_003.png",
    "readme.txt",
    "data_file.jpg",
]

for filename in test_files:
    filepath = os.path.join(temp_dir, filename)
    with open(filepath, 'w') as f:
        f.write("test content")

# Define ARGS for load_folder function
ARGS = {
    "folder": temp_dir,
    "suffix": ".jpg"
}
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 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
load_imageid Ran
ran on a synthesized input
jnhwkim/ban-vqa/utils.py
code served (permissive licence) · get_code("bf90bd27fe08f3a2")
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 set. 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 a temporary directory with test image files
temp_dir = tempfile.mkdtemp()

# Create some test jpg files with names matching the expected pattern
test_files = [
    "image_001.jpg",
    "image_002.jpg",
    "image_003.jpg",
    "other_file_001.jpg",
]

for fname in test_files:
    filepath = os.path.join(temp_dir, fname)
    with open(filepath, 'w') as f:
        f.write("dummy image data")

# Define arguments for load_imageid function
ARGS = {
    "folder": temp_dir
}
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 set — a set output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a set 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
pil_loader Ran
ran, honoured the contract we drafted
jnhwkim/ban-vqa/utils.py
pointer only (licence: MIT); the stored copy get_code serves, from YuemingJin/MTRCNet-CL, is pointer only (licence: NONE) · get_code("1df9a5ffd9b38c34")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from YuemingJin/MTRCNet-CL, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

prepare_answers Ran
ran on a synthesized input
Cyanogenoid/vqa-counting/vqa-v2/data.py
code served (permissive licence) · get_code("1e3e251c3f07251d")
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 generator. 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 re

_period_strip = re.compile('(?!<=\\d)(\\.)(?!\\d)')

_comma_strip = re.compile('(\\d)(,)(\\d)')

_punctuation_chars = re.escape(';/[]"{}()=+\\_-><@`,?!')

_punctuation = re.compile('([{}])'.format(re.escape(_punctuation_chars)))

_punctuation_with_a_space = re.compile('(?<= )([{0}])|([{0}])(?= )'.format(_punctuation_chars))

def prepare_answers(answers_json):
    """ Normalize answers from a given answer json in the usual VQA format. """
    answers = [[a['answer'] for a in ans_dict['answers']] for ans_dict in answers_json['annotations']]

    def process_punctuation(s):
        if _punctuation.search(s) is None:
            return s
        s = _punctuation_with_a_space.sub('', s)
        if re.search(_comma_strip, s) is not None:
            s = s.replace(',', '')
        s = _punctuation.sub(' ', s)
        s = _period_strip.sub('', s)
        return s.strip()
    for answer_list in answers:
        yield list(map(process_punctuation, answer_list))


answers_json = {
    'annotations': [
        {
            'answers': [
                {'answer': 'yes'},
                {'answer': 'no, maybe'}
            ]
        },
        {
            'answers': [
                {'answer': 'a dog; running'},
                {'answer': '123,456'}
            ]
        }
    ]
}

ARGS = {
    'answers_json': answers_json
}
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 generator — a generator output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a generator 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
tile_2d_over_nd Ran
ran, violated the contract we drafted
Cyanogenoid/vqa-counting/vqa-v2/model.py
code served (permissive licence) · get_code("b3299f7858d7d2f2")
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, 4, 3, 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)

# feature_vector: a 2D tensor of shape (batch_size, num_features)
# feature_map: a tensor with shape (batch_size, num_features, *spatial_sizes)
# The function will repeat feature_vector across all spatial positions

feature_vector = torch.randn(2, 4)  # batch_size=2, num_features=4

# feature_map should have batch_size=2, num_features=4, and some spatial dimensions
# Let's use spatial dimensions of 3x5 to be distinct
feature_map = torch.randn(2, 4, 3, 5)

ARGS = {
    "feature_vector": feature_vector,
    "feature_map": feature_map,
}
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, 4, 3, 5) — shape (2, 4, 3, 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

Also present in other papers' code

Of 13 samples lifted from this paper's linked repositories, 13 carry a function identity and 0 do not. Of the 13 distinct functions with an identity, 7 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 7 later) and 6 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
pil_loader 28 other papers
1 dated earlier, 27 later
1702.00758 (2017-02)
HashNet: Deep Learning to Hash by Continuation
1807.06587 (2018-07)
Deep Exemplar-based Colorization
1902.09368 (2019-02)
Dual Attention Networks for Visual Reference Resolution in Visual Dialog
and 25 more not named here
load_folder 7 other papers
7 later
1902.09368 (2019-02)
Dual Attention Networks for Visual Reference Resolution in Visual Dialog
2006.14744 (2020-06)
Graph Optimal Transport for Cross-Domain Alignment
2108.04024 (2021-08)
Image Retrieval on Real-life Images with Pre-trained Vision-and-Language Models
and 4 more not named here
answer_filter 5 other papers
5 later
1903.12314 (2019-03)
Relation-Aware Graph Attention Network for Visual Question Answering
1909.11874 (2019-09)
Compact Trilinear Interaction for Visual Question Answering
2003.10286 (2020-03)
PathVQA: 30000+ Questions for Medical Visual Question Answering
and 2 more not named here
is_howmany 4 other papers
4 later
1903.12314 (2019-03)
Relation-Aware Graph Attention Network for Visual Question Answering
2003.10286 (2020-03)
PathVQA: 30000+ Questions for Medical Visual Question Answering
2006.14744 (2020-06)
Graph Optimal Transport for Cross-Domain Alignment
and 1 more not named here
load_imageid 4 other papers
4 later
1902.09368 (2019-02)
Dual Attention Networks for Visual Reference Resolution in Visual Dialog
2006.14744 (2020-06)
Graph Optimal Transport for Cross-Domain Alignment
2110.02526 (2021-10)
Coarse-to-Fine Reasoning for Visual Question Answering
and 1 more not named here
compute_recall_with_logits 1 other paper
1 later
2006.14744 (2020-06)
Graph Optimal Transport for Cross-Domain Alignment
tile_2d_over_nd 1 other paper
1 later
1812.01880 (2018-12)
Learning to Compose Dynamic Tree Structures for Visual Contexts

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

Attention networks in multimodal learning provide an efficient way to utilize given visual information selectively. However, the computational cost to learn attention distributions for every pair of multimodal input channels is prohibitively expensive. To solve this problem, co-attention builds two separate attention distributions for each modality neglecting the interaction between multimodal inputs. In this paper, we propose bilinear attention networks (BAN) that find bilinear attention distributions to utilize given vision-language information seamlessly. BAN considers bilinear interactions among two groups of input channels, while low-rank bilinear pooling extracts the joint representations for each pair of channels. Furthermore, we propose a variant of multimodal residual networks to exploit eight-attention maps of the BAN efficiently. We quantitatively and qualitatively evaluate our model on visual question answering (VQA 2.0) and Flickr30k Entities datasets, showing that BAN significantly outperforms previous methods and achieves new state-of-the-arts on both datasets.

For agents

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

get_harvested_code_for_paper("1805.07932")
get_code_for_paper("1805.07932")
have("1805.07932")

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

curl https://syntology.ai/api/ran/1805.07932.json

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

[![Syntology run record](https://syntology.ai/api/ran/1805.07932.svg)](https://syntology.ai/paper/1805.07932)

Connect an agent — have() is free.