SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · aaai_28059 · AAAI · 2024

Gaussian Mixture Proposals with Pull-Push Learning Scheme to Capture Diverse Events for Weakly Supervised Temporal Video Grounding

Youngjoon Yoo, Jin Choi, Jungchan Cho, Sunoh Kim, Joonsang Yu

Open in the Atlas

Code that ran

6 of 8 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 5 have no contract verdict; 0 failures of our instrument; 2 no recorded run), measured 2026-11-22; 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/aaai_28059.json · badge: /api/ran/aaai_28059.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
sunoh-kim/pps canonical (classifier label)
classifier basis: census
6 ran · 2 no recorded run
FunctionStatusWhere it lives
build_collate_data Ran
ran on a synthesized input
sunoh-kim/pps/dataset/base.py
code served (permissive licence) · get_code("cb21b39e657b4027")
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 function. 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
import torch

torch.manual_seed(0)

# Parameters for build_collate_data
max_num_segments = 5
max_num_words = 10
frame_dim = 4
word_dim = 3

# Create sample data that matches what the collate_data function expects
sample_1 = {
    'raw': 'sample text 1',
    'frames_feat': np.random.randn(3, 4).astype(np.float32),
    'words_feat': np.random.randn(7, 3).astype(np.float32),
    'words_id': np.array([1, 2, 3, 4, 5, 6]).astype(np.int64),
    'weights': np.random.randn(6).astype(np.float32),
}

sample_2 = {
    'raw': 'sample text 2',
    'frames_feat': np.random.randn(4, 4).astype(np.float32),
    'words_feat': np.random.randn(8, 3).astype(np.float32),
    'words_id': np.array([7, 8, 9, 10, 11]).astype(np.int64),
    'weights': np.random.randn(5).astype(np.float32),
}

samples = [sample_1, sample_2]

ARGS = {
    'max_num_segments': max_num_segments,
    'max_num_words': max_num_words,
    'frame_dim': frame_dim,
    'word_dim': word_dim,
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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
cal_nll_loss Ran
ran on a synthesized input
sunoh-kim/pps/model/loss.py
code served (permissive licence) · get_code("4e5b6a1ac60347a3")
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 torch.nn as nn

torch.manual_seed(0)

# Small batch size: 2
# Sequence length: 3
# Number of classes: 4

batch_size = 2
seq_length = 3
num_classes = 4

# logit: shape (batch_size, seq_length, num_classes)
# Output of a model before softmax
logit = torch.randn(batch_size, seq_length, num_classes)

# idx: shape (batch_size, seq_length)
# Target class indices
idx = torch.randint(0, num_classes, (batch_size, seq_length))

# mask: shape (batch_size, seq_length)
# Binary mask for valid positions
mask = torch.ones(batch_size, seq_length)
mask[0, 2] = 0  # Mask out one position to test masking

# weights: optional, shape (batch_size, seq_length)
# Per-position weights
weights = torch.ones(batch_size, seq_length) * 0.5

ARGS = {
    "logit": logit,
    "idx": idx,
    "mask": mask,
    "weights": weights,
}
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
calculate_IoU Ran
ran on a synthesized input
sunoh-kim/pps/runner/base_runner.py
code served (permissive licence) · get_code("e86c4a293d3aaad6")
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 1, dtype family float, shape (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

# Analyze the function: calculate_IoU takes two arguments i0 and i1
# Each appears to be a tuple of two numpy arrays (start and end points)
# The function performs interval union/intersection operations

# Create two intervals as tuples of (start_array, end_array)
# The arrays represent boundaries that can be compared element-wise

i0_start = np.array([1.0, 2.0, 3.0, 4.0])
i0_end = np.array([5.0, 6.0, 7.0, 8.0])
i0 = (i0_start, i0_end)

i1_start = np.array([2.0, 3.0, 4.0, 5.0])
i1_end = np.array([6.0, 7.0, 8.0, 9.0])
i1 = (i1_start, i1_end)

ARGS = {
    "i0": i0,
    "i1": i1,
}
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 1, dtype family float, shape (4,) — shape (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
fill_with_neg_inf Ran
ran, honoured the contract we drafted
sunoh-kim/pps/model/module/mutihead_attention.py
pointer only (licence: MIT); the stored copy get_code serves, from kenford953/graphcage, is pointer only (licence: NONE) · get_code("8f266b9d616de9b9")
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 kenford953/graphcage, 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.

top_1_metric Ran
ran on a synthesized input
sunoh-kim/pps/runner/base_runner.py
code served (permissive licence) · get_code("c9b78ff2994ac2de")
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 numpy as np

# Analyze the function:
# - top_1_metric takes two parameters: pred and label
# - Both are accessed with [:, 0] and [:, 1], so they need shape (batch_size, 2)
# - calculate_IoU expects tuples of arrays with shape (batch_size,)
# - The function computes metrics on these predictions and labels

# Create small test data
batch_size = 2

# pred and label should be numpy arrays with shape (batch_size, 2)
# Representing intervals/ranges with start and end values
pred = np.array([
    [1.0, 5.0],
    [2.0, 7.0]
], dtype=np.float32)

label = np.array([
    [1.5, 6.0],
    [2.5, 8.0]
], dtype=np.float32)

ARGS = {
    'pred': pred,
    'label': label
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
top_n_metric Ran
ran on a synthesized input
sunoh-kim/pps/runner/base_runner.py
code served (permissive licence) · get_code("28409e74ced69fb1")
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 numpy as np

# Analyze the function:
# - preds: list of arrays, each with shape (bsz, 2) where [:, 0] and [:, 1] are interval endpoints
# - label: array with shape (bsz, 2) where [:, 0] and [:, 1] are interval endpoints
# - bsz is extracted from preds[0].shape[0]

# Create small sample data
np.random.seed(0)

# Batch size
bsz = 2

# Create label: (bsz, 2) array with interval start and end points
# Start points should be less than end points
label = np.array([
    [1.0, 5.0],   # interval 1
    [2.0, 7.0]    # interval 2
], dtype=np.float32)

# Create preds: list of predictions, each is (bsz, 2)
# Multiple predictions to test the "top-n" aspect
pred1 = np.array([
    [1.5, 4.5],   # close to label[0]
    [2.5, 6.5]    # close to label[1]
], dtype=np.float32)

pred2 = np.array([
    [0.5, 5.5],   # another prediction for sample 0
    [1.5, 7.5]    # another prediction for sample 1
], dtype=np.float32)

pred3 = np.array([
    [1.2, 4.8],   # third prediction for sample 0
    [2.2, 6.8]    # third prediction for sample 1
], dtype=np.float32)

preds = [pred1, pred2, pred3]

ARGS = {
    'preds': preds,
    'label': label
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
main_loss_fn Not yet run
unverified (harvested, no recorded run)
sunoh-kim/pps/model/loss.py
code served (permissive licence) · get_code("ca30cf07c90f45aa")
How it ran

no invocation recorded.

sub_loss_fn Not yet run
unverified (harvested, no recorded run)
sunoh-kim/pps/model/loss.py
code served (permissive licence) · get_code("7ca187567f20bbd0")
How it ran

no invocation recorded.

Also present in other papers' code

Of 8 samples lifted from this paper's linked repositories, 8 carry a function identity and 0 do not. Of the 8 distinct functions with an identity, 4 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 4 later) and 4 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
fill_with_neg_inf 16 other papers
9 dated earlier, 2 later, 5 same year, order not determinable · 3 only as an assembly brick
1906.00295 (2019-06)
Multimodal Transformer for Unaligned Multimodal Language Sequences
2007.00808 (2020-07)
Approximate Nearest Neighbor Negative Contrastive Learning for Dense Text Retrieval
2109.04096 (2021-09)
A Three-Stage Learning Framework for Low-Resource Knowledge-Grounded Dialogue Generation
and 13 more not named here
cal_nll_loss 1 other paper
1 later
2602.03071 (2026-02)
Boundary Optimization for Weakly Supervised Video Grounding
main_loss_fn 1 other paper
1 later
2602.03071 (2026-02)
Boundary Optimization for Weakly Supervised Video Grounding
sub_loss_fn 1 other paper
1 later
2602.03071 (2026-02)
Boundary Optimization for Weakly Supervised Video Grounding

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

In the weakly supervised temporal video grounding study, previous methods use predetermined single Gaussian proposals which lack the ability to express diverse events described by the sentence query. To enhance the expression ability of a proposal, we propose a Gaussian mixture proposal (GMP) that can depict arbitrary shapes by learning importance, centroid, and range of every Gaussian in the mixture. In learning GMP, each Gaussian is not trained in a feature space but is implemented over a temporal location. Thus the conventional feature-based learning for Gaussian mixture model is not valid for our case. In our special setting, to learn moderately coupled Gaussian mixture capturing diverse events, we newly propose a pull-push learning scheme using pulling and pushing losses, each of which plays an opposite role to the other. The effects of components in our scheme are verified in-depth with extensive ablation studies and the overall scheme achieves state-of-the-art performance. Our code is available at https://github.com/sunoh-kim/pps.

For agents

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

get_harvested_code_for_paper("aaai_28059")
get_code_for_paper("aaai_28059")
have("aaai_28059")

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

curl https://syntology.ai/api/ran/aaai_28059.json

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

[![Syntology run record](https://syntology.ai/api/ran/aaai_28059.svg)](https://syntology.ai/paper/aaai_28059)

Connect an agent — have() is free.