SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2008.04237 · 2020

Self-Supervised Learning of Audio-Visual Objects from Video

arXiv · PDF · Open in the Atlas

Code that ran

11 of 14 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; 0 honoured the contract we drafted, 0 violated it and 10 have no contract verdict; 1 failure of our instrument; 3 no recorded run), measured 2026-11-21; 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/2008.04237.json · badge: /api/ran/2008.04237.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
afourast/avobjects community (classifier label)
classifier basis not recorded on the link
11 ran · 3 no recorded run
FunctionStatusWhere it lives
Backward Ran
ran on a synthesized input
afourast/avobjects/flow/pwcnet.py
code served (permissive licence) · get_code("69a7d0834579bdbd")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float, shape (2, 3, 4, 5) as a later run recorded it. The image it ran in was not recorded. 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, 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 as nn

torch.manual_seed(0)

# The function expects:
# - tensorInput: a 4D tensor (batch, channels, height, width)
# - tensorFlow: a 4D tensor (batch, 2, height, width) representing optical flow

# Create small tensors with distinct dimensions for clarity
batch_size = 2
input_channels = 3
height = 4
width = 5
flow_channels = 2

# tensorInput: (batch, channels, height, width)
tensorInput = torch.randn(batch_size, input_channels, height, width)

# tensorFlow: (batch, 2, height, width) - 2 channels for horizontal and vertical flow
tensorFlow = torch.randn(batch_size, flow_channels, height, width)

ARGS = {
    'tensorInput': tensorInput,
    'tensorFlow': tensorFlow,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 3, 4, 5) — shape (2, 3, 4, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
calc_receptive_field Ran
ran on a synthesized input
afourast/avobjects/utils.py
code served (permissive licence) · get_code("fef192e75c60a68c")
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 math

def outFromIn(conv, layerIn, ceil_mode=True):
    n_in = layerIn[0]
    j_in = layerIn[1]
    r_in = layerIn[2]
    start_in = layerIn[3]
    k = conv[0]
    s = conv[1]
    p = conv[2]
    n_out = math.floor((n_in - k + 2 * p) / s) + 1
    actualP = (n_out - 1) * s - n_in + k
    pR = math.ceil(actualP / 2)
    pL = math.floor(actualP / 2)
    j_out = j_in * s
    r_out = r_in + (k - 1) * j_in
    start_out = start_in + ((k - 1) / 2 - pL) * j_in
    return (n_out, j_out, r_out, start_out)

def calc_receptive_field(layers, imsize, layer_names=None):
    if layer_names is not None:
        print('-------Net summary------')
    currentLayer = [imsize, 1, 1, 0.5]
    for l_id, layer in enumerate(layers):
        conv = [layer[key][-1] if type(layer[key]) in [list, tuple] else layer[key] for key in ['kernel_size', 'stride', 'padding']]
        currentLayer = outFromIn(conv, currentLayer)
        if 'maxpool' in layer:
            conv = [(layer['maxpool'][key][-1] if type(layer['maxpool'][key]) in [list, tuple] else layer['maxpool'][key]) if not key == 'padding' or 'padding' in layer['maxpool'] else 0 for key in ['kernel_size', 'stride', 'padding']]
            currentLayer = outFromIn(conv, currentLayer, ceil_mode=False)
    return currentLayer

ARGS = {
    'layers': [
        {
            'kernel_size': 3,
            'stride': 1,
            'padding': 1,
        },
        {
            'kernel_size': 3,
            'stride': 1,
            'padding': 1,
            'maxpool': {
                'kernel_size': 2,
                'stride': 2,
                'padding': 0,
            }
        },
        {
            'kernel_size': (3, 3),
            'stride': (2, 2),
            'padding': (1, 1),
        }
    ],
    'imsize': 224,
    'layer_names': ['conv1', 'conv2_maxpool', 'conv3']
}
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 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
calc_up_down_strides Ran
ran on a synthesized input
afourast/avobjects/sepnet.py
code served (permissive licence) · get_code("9fb9b2b48e146b7a")
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 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

# The function calc_up_down_strides takes two parameters:
# 1. num_layers: an integer representing total number of layers
# 2. stride: a number (int or float) representing the stride value (2 for downsampling, 0.5 for upsampling)

# Looking at the function body:
# - num_layers is used in arithmetic operations and list creation
# - stride is used as a value to assign to list elements
# No complex objects needed, just simple types.

ARGS = {
    "num_layers": 5,
    "stride": 2
}
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 1, dtype family int — 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
gpu_initializer Ran
ran on a synthesized input
afourast/avobjects/utils.py
code served (permissive licence) · get_code("67ccf701131b0af7")
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 device. 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 os

torch.manual_seed(0)

# The function gpu_initializer takes a gpu_id parameter
# It uses it to set CUDA_VISIBLE_DEVICES environment variable
# A valid gpu_id would be an integer or string representing a GPU device number

ARGS = {
    'gpu_id': 0
}
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 device — a device output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a device 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
load_wav Ran
ran on a synthesized input
afourast/avobjects/load_audio.py
code served (permissive licence) · get_code("33b1b8298f952f5b")
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 (16000,) 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
from scipy.io import wavfile
import tempfile
import os

# Create a temporary WAV file for testing
temp_dir = tempfile.mkdtemp()
temp_wav_path = os.path.join(temp_dir, "test_audio.wav")

# Generate a simple audio signal
sample_rate = 16000
duration = 2  # 2 seconds
t = np.linspace(0, duration, int(sample_rate * duration))
frequency = 440  # A4 note
audio_data = np.sin(2 * np.pi * frequency * t).astype(np.float32)

# Write the WAV file
wavfile.write(temp_wav_path, sample_rate, audio_data)

# Define the ARGS dictionary
ARGS = {
    'path': temp_wav_path,
    'fr': 0,
    'to': 1,
    'sample_rate': 16000
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — 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 (16000,) — shape (16000,) 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
normalize_img Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
afourast/avobjects/viz_utils.py
code served (permissive licence) · get_code("e7357eb51f7bea31")
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, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create a small tensor for normalization
value = torch.tensor([[1.0, 2.0, 3.0, 4.0],
                      [5.0, 6.0, 7.0, 8.0]])

# Set up ARGS with all parameters
ARGS = {
    'value': value,
    'vmax': None,
    'vmin': None
}
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, 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
outFromIn Ran
ran on a synthesized input
afourast/avobjects/utils.py
code served (permissive licence) · get_code("616702d60caf4dcd")
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 math

def outFromIn(conv, layerIn, ceil_mode=True):
    n_in = layerIn[0]
    j_in = layerIn[1]
    r_in = layerIn[2]
    start_in = layerIn[3]
    k = conv[0]
    s = conv[1]
    p = conv[2]
    n_out = math.floor((n_in - k + 2 * p) / s) + 1
    actualP = (n_out - 1) * s - n_in + k
    pR = math.ceil(actualP / 2)
    pL = math.floor(actualP / 2)
    j_out = j_in * s
    r_out = r_in + (k - 1) * j_in
    start_out = start_in + ((k - 1) / 2 - pL) * j_in
    return (n_out, j_out, r_out, start_out)

# conv is a tuple/list of (kernel_size, stride, padding)
conv = (3, 2, 1)

# layerIn is a tuple/list of (n_in, j_in, r_in, start_in)
# These represent input feature map size, jump, receptive field, start position
layerIn = (28, 1, 1, 0.5)

# ceil_mode is a boolean (not actually used in the function body)
ceil_mode = True

ARGS = {
    "conv": conv,
    "layerIn": layerIn,
    "ceil_mode": ceil_mode
}
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 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
relative_module Ran
ran on a synthesized input
afourast/avobjects/aolib_p3/areload.py
code served (permissive licence) · get_code("695621820eee0aee")
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 bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import types

# Create a mock module with __file__ attribute set to a relative path
mock_module_relative = types.ModuleType('mock_relative')
mock_module_relative.__file__ = 'relative/path/to/module.py'

ARGS = {
    'm': mock_module_relative
}
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 bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
tensor_from_im Ran
ran on a synthesized input
afourast/avobjects/flow/pwcnet.py
code served (permissive licence) · get_code("64bc62b1a123906a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 3, dtype family float, shape (3, 4, 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
import torch

torch.manual_seed(0)

# The function expects 'im' which should be a numpy array
# Looking at the function body:
# - im[:, :, ::-1] reverses the last dimension (color channel reversal)
# - .transpose(2, 0, 1) moves channels to front
# - .astype(np.float32) converts to float32
# - Multiplies by 1.0/255.0 to normalize
# So 'im' should be a 3D numpy array (height, width, channels) with uint8 values

# Create a small test image: 4 height x 3 width x 3 channels (RGB)
im = np.random.randint(0, 256, size=(4, 3, 3), dtype=np.uint8)

ARGS = {
    "im": im
}
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 Tensor, ndim 3, dtype family float, shape (3, 4, 3) — shape (3, 4, 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
torch_mag_phase_2_np_complex Ran
ran on a synthesized input
afourast/avobjects/load_audio.py
code served (permissive licence) · get_code("0b0f46b99a26aeb4")
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 (4, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Create small magnitude and phase tensors
# mag_spect: magnitude spectrogram - typical shape would be (freq_bins, time_steps)
# phase: phase spectrogram - same shape as mag_spect
mag_spect = torch.rand(4, 3)  # (freq_bins=4, time_steps=3)
phase = torch.rand(4, 3) * 2 * np.pi  # (freq_bins=4, time_steps=3), phase in [0, 2π]

ARGS = {
    "mag_spect": mag_spect,
    "phase": phase,
}
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 ndarray, ndim 2, dtype family float, shape (4, 3) — shape (4, 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
viz_boxes_with_scores Ran
ran on a synthesized input
afourast/avobjects/viz_utils.py
code served (permissive licence) · get_code("71c1d1306fe15964")
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 4, dtype family int, shape (3, 3, 64, 80) 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

# Small mock for aolib_p3 module since it's imported by the function
class MockImg:
    @staticmethod
    def colors_from_input(colors, default, n):
        if colors is None:
            return [default] * n
        return colors
    
    @staticmethod
    def draw_rects(im, rects, fills, outlines, texts, line_widths):
        return im
    
    @staticmethod
    def draw_text(im, texts, text_loc, colors, font_size):
        return im

class MockUtil:
    @staticmethod
    def distinct_colors(n):
        return [(255, 0, 0) for _ in range(n)]

# Mock the aolib_p3 module
import sys
from types import ModuleType

aolib_p3 = ModuleType('aolib_p3')
aolib_p3.img = MockImg()
aolib_p3.util = MockUtil()

img_module = ModuleType('img')
util_module = ModuleType('util')
img_module.colors_from_input = MockImg.colors_from_input
img_module.draw_rects = MockImg.draw_rects
img_module.draw_text = MockImg.draw_text
util_module.distinct_colors = MockUtil.distinct_colors

aolib_p3.img = img_module
aolib_p3.util = util_module

sys.modules['aolib_p3'] = aolib_p3
sys.modules['aolib_p3.img'] = img_module
sys.modules['aolib_p3.util'] = util_module

# Construct arguments
# video: t h w c format
video = np.random.randint(0, 256, size=(3, 64, 80, 3), dtype=np.uint8)

# box_centers: either (n_boxes, t, 2) for centers only, or (n_boxes, t, 4) for centers + sizes
# Using (n_boxes, t, 2) format - will trigger add_cont_bb_size_to_traj
box_centers = np.array([
    [[10.5, 20.3], [15.2, 22.1], [18.0, 25.5]],
    [[50.1, 40.2], [52.3, 42.5], [54.1, 45.0]]
], dtype=np.float32)

# scores: (n_boxes, t) shape
scores = np.array([
    [0.95, 0.87, 0.92],
    [0.78, 0.82, 0.88]
], dtype=np.float32)

# const_box_size: scalar
const_box_size = 16

# colors: list of RGB tuples or None (will use distinct_colors)
colors = None

# asd_thresh: threshold for filtering boxes by score
asd_thresh = 0.8

ARGS = {
    'video': video,
    'box_centers': box_centers,
    'scores': scores,
    'const_box_size': const_box_size,
    'colors': colors,
    'asd_thresh': asd_thresh
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
aolib_p3, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 4, dtype family int, shape (3, 3, 64, 80) — shape (3, 3, 64, 80) 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
estimate Not yet run
unverified (harvested, no recorded run)
afourast/avobjects/flow/pwcnet.py
code served (permissive licence) · get_code("0747b9a053d42ac4")
How it ran

no invocation recorded.

load_mp4 Not yet run
unverified (harvested, no recorded run)
afourast/avobjects/load_video.py
code served (permissive licence) · get_code("e71c25d54b32ce0e")
How it ran

no invocation recorded.

load_mp4_ffmpeg Not yet run
unverified (harvested, no recorded run)
afourast/avobjects/load_video.py
code served (permissive licence) · get_code("df22707a0588254b")
How it ran

no invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 distinct functions with an identity, 1 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 1 later) and 13 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
normalize_img 2 other papers
2 later
2209.09634 (2022-09)
A Closer Look at Weakly-Supervised Audio-Visual Source Localization
2309.05281 (2023-09)
Class-Incremental Grouping Network for Continual Audio-Visual Learning

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 2026-11-21.

Repositories linked to this paper

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

Abstract

Our objective is to transform a video into a set of discrete audio-visual objects using self-supervised learning. To this end, we introduce a model that uses attention to localize and group sound sources, and optical flow to aggregate information over time. We demonstrate the effectiveness of the audio-visual object embeddings that our model learns by using them for four downstream speech-oriented tasks: (a) multi-speaker sound source separation, (b) localizing and tracking speakers, (c) correcting misaligned audio-visual data, and (d) active speaker detection. Using our representation, these tasks can be solved entirely by training on unlabeled video, without the aid of object detectors. We also demonstrate the generality of our method by applying it to non-human speakers, including cartoons and puppets.Our model significantly outperforms other self-supervised approaches, and obtains performance competitive with methods that use supervised face detection.

For agents

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

get_harvested_code_for_paper("2008.04237")
get_code_for_paper("2008.04237")
have("2008.04237")

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

curl https://syntology.ai/api/ran/2008.04237.json

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

[![Syntology run record](https://syntology.ai/api/ran/2008.04237.svg)](https://syntology.ai/paper/2008.04237)

Connect an agent — have() is free.