SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1812.05038 · 2018

Long-Term Feature Banks for Detailed Video Understanding

arXiv · PDF · Open in the Atlas

Code that ran

8 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; 0 honoured the contract we drafted, 0 violated it and 8 have no contract verdict; 0 failures 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/1812.05038.json · badge: /api/ran/1812.05038.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
facebookresearch/video-long-term-feature-banks canonical (classifier label)
classifier basis: census
8 ran · 0 no recorded run
FunctionStatusWhere it lives
aggregate_labels Ran
ran on a synthesized input
facebookresearch/video-long-term-feature-banks/lib/datasets/charades.py
code served (permissive licence) · get_code("a2a4e1ac0aaddbce")
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 torch
import torch.nn as nn

torch.manual_seed(0)

# Analyze the function:
# - It takes `label_list` as parameter
# - It iterates over `label_list` (outer loop)
# - For each element `labels` in `label_list`, it iterates over it (inner loop)
# - It appends each element `l` to `all_labels`
# - It returns the unique set of all labels
# 
# So `label_list` should be a sequence of sequences (list of lists)
# Each inner list contains labels (can be strings, ints, etc.)

# Create a simple label_list with nested lists of labels
label_list = [
    [1, 2, 3],
    [2, 3, 4],
    [1, 5]
]

ARGS = {
    "label_list": label_list
}
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 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
convert_to_batch Ran
ran on a synthesized input
facebookresearch/video-long-term-feature-banks/lib/datasets/data_input_helper.py
code served (permissive licence) · get_code("2a790105b9c533f5")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# The function expects 'data' to be an iterable of array-like objects
# It will concatenate them along a new axis and convert to float32
# Let's create a list of small numpy arrays

arr1 = np.array([1.0, 2.0, 3.0, 4.0])
arr2 = np.array([5.0, 6.0, 7.0, 8.0])

data = [arr1, arr2]

ARGS = {
    "data": data
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_sequence Ran
ran on a synthesized input
facebookresearch/video-long-term-feature-banks/lib/datasets/dataset_helper.py
code served (permissive licence) · get_code("1c2aad8821b0a87f")
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 range. 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 numpy as np

torch.manual_seed(0)

ARGS = {
    "center_idx": 50,
    "half_len": 10,
    "sample_rate": 2,
    "num_frames": 100
}
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 range — a range output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a range output the recorded ndim 1 and dtype family int 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_boxes_and_labels Ran
ran on a synthesized input
facebookresearch/video-long-term-feature-banks/lib/datasets/ava.py
code served (permissive licence) · get_code("808883fafe444db8")
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 tempfile
import os

# Create temporary CSV files for testing
temp_dir = tempfile.mkdtemp()

# Create a training CSV file
train_csv = os.path.join(temp_dir, "train.csv")
with open(train_csv, 'w') as f:
    # Format: video_name, frame_sec, x1, y1, x2, y2, label, [score]
    f.write("video1,902,0.1,0.2,0.5,0.6,1\n")
    f.write("video1,902,0.2,0.3,0.6,0.7,2\n")
    f.write("video1,906,0.3,0.4,0.7,0.8,3\n")
    f.write("video1,910,0.4,0.5,0.8,0.9,-1\n")
    f.write("video2,902,0.15,0.25,0.55,0.65,1\n")

# Create an inference CSV file with detection scores
infer_csv = os.path.join(temp_dir, "infer.csv")
with open(infer_csv, 'w') as f:
    # Format with score: video_name, frame_sec, x1, y1, x2, y2, label, score
    f.write("video1,902,0.1,0.2,0.5,0.6,-1,0.9\n")
    f.write("video1,906,0.25,0.35,0.65,0.75,-1,0.85\n")
    f.write("video1,910,0.3,0.4,0.7,0.8,-1,0.5\n")
    f.write("video2,902,0.15,0.25,0.55,0.65,-1,0.95\n")

ARGS = {
    "filenames": [train_csv],
    "is_train": True,
    "detect_thresh": 0.0,
    "full_eval": False,
}
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
retry_load_images Ran
ran on a synthesized input
facebookresearch/video-long-term-feature-banks/lib/datasets/data_input_helper.py
code served (permissive licence) · get_code("3cac88e08ef2456a")
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 4, 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 cv2
import numpy as np
import tempfile
import os

# Create temporary test images
temp_dir = tempfile.mkdtemp()

# Create two valid test images
img1 = np.zeros((10, 10, 3), dtype=np.uint8)
img1[0:5, 0:5] = [255, 0, 0]  # Red square
img2 = np.ones((10, 10, 3), dtype=np.uint8) * 128
img2[0:5, 0:5] = [0, 255, 0]  # Green square

# Write images to disk
path1 = os.path.join(temp_dir, "test_image1.png")
path2 = os.path.join(temp_dir, "test_image2.png")
cv2.imwrite(path1, img1)
cv2.imwrite(path2, img2)

ARGS = {
    "image_paths": [path1, path2],
    "retry": 2
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 4, 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
sample_center_of_segments Ran
ran on a synthesized input
facebookresearch/video-long-term-feature-banks/lib/datasets/charades.py
code served (permissive licence) · get_code("66aa5a4540720019")
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 int, ndim 0, dtype family int, 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 numpy as np

# Analyzing the function:
# - segment_id: an integer index into segments
# - num_frames: total number of frames (integer)
# - num_test_segments: number of segments to divide video into (integer)
# - half_len: parameter that exists but is NOT used in the function body

ARGS = {
    "segment_id": 2,
    "num_frames": 100,
    "num_test_segments": 10,
    "half_len": 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 int, ndim 0, dtype family int, 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
sample_train_idx Ran
ran on a synthesized input
facebookresearch/video-long-term-feature-banks/lib/datasets/charades.py
code served (permissive licence) · get_code("869476ae19df5c8b")
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 int, ndim 0, dtype family int, 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 random

def sample_train_idx(num_frames, seq_len):
    """Sample training frames."""
    half_len = seq_len // 2
    if num_frames < seq_len:
        center_idx = num_frames // 2
    else:
        center_idx = random.randint(half_len, num_frames - half_len)
    return center_idx

random.seed(42)

ARGS = {
    "num_frames": 100,
    "seq_len": 16
}
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 int, ndim 0, dtype family int, 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
time_to_sec Ran
ran on a synthesized input
facebookresearch/video-long-term-feature-banks/lib/datasets/epic.py
code served (permissive licence) · get_code("45fec9a5e6665aa8")
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 float, 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 torch
import numpy as np

# Analyze the function: time_to_sec(sec)
# - Takes a parameter 'sec' which is a string
# - The string is expected to be in format "HH:MM:SS.ss"
# - It splits by ':' and expects 3 parts: hour, minute, sec
# - Returns a float representing total seconds

ARGS = {
    'sec': '00:02:10.99'
}
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 float, 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

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, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 8 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

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

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

Repositories linked to this paper

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

Abstract

To understand the world, we humans constantly need to relate the present to the past, and put events in context. In this paper, we enable existing video models to do the same. We propose a long-term feature bank---supportive information extracted over the entire span of a video---to augment state-of-the-art video models that otherwise would only view short clips of 2-5 seconds. Our experiments demonstrate that augmenting 3D convolutional networks with a long-term feature bank yields state-of-the-art results on three challenging video datasets: AVA, EPIC-Kitchens, and Charades.

For agents

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

get_harvested_code_for_paper("1812.05038")
get_code_for_paper("1812.05038")
have("1812.05038")

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

curl https://syntology.ai/api/ran/1812.05038.json

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

[![Syntology run record](https://syntology.ai/api/ran/1812.05038.svg)](https://syntology.ai/paper/1812.05038)

Connect an agent — have() is free.