SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2206.01720 · CVPR · 2022

Revisiting the "Video" in Video-Language Understanding

Jiajun Wu, Adrien Gaidon, Li Fei-Fei, Juan Niebles, Cristóbal Eyzaguirre, Shyamal Buch

arXiv · PDF · Open in the Atlas

Code that ran

3 of 5 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (2 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 3 have no contract verdict; 0 failures of our instrument; 2 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/2206.01720.json · badge: /api/ran/2206.01720.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
stanfordvl/atp-video-language community
Papers with Code archive
3 ran · 2 no recorded run
FunctionStatusWhere it lives
ATPConfig Ran
ran on a synthesized input · constructed an object; no result was computed
stanfordvl/atp-video-language/src/atp.py
code served (permissive licence) · get_code("651caa45e12996b1")
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 an instance of its own class, ATPConfig: the run constructed the object and did not call it, so no result was computed. 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
from dataclasses import dataclass

torch.manual_seed(0)

ARGS = {
    "n_layers": 2,
    "n_heads": 2,
    "d_model": 128,
    "d_model_ff": 128,
    "enc_dropout": 0.1,
    "use_text_query": False,
    "use_text_cands": False,
    "n_cands": 5,
    "use_ste": True,
    "sel_dropout": 0.0,
    "d_input": 512,
}
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 ATPConfig — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ATPConfig 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
ATPEncoder Ran
ran on a synthesized input
stanfordvl/atp-video-language/src/atp.py
code served (permissive licence) · get_code("5cdaefd6e6a40813")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
from torch import nn
from dataclasses import dataclass
from enum import IntEnum
import math

@dataclass
class ATPConfig:
    """
    ATPConfig contains the parameters needed for the ATPSelectorModel (and its ATPEncoder).
    """
    n_layers: int = 2
    n_heads: int = 2
    d_model: int = 128
    d_model_ff: int = 128
    enc_dropout: float = 0.1
    use_text_query: bool = False
    use_text_cands: bool = False
    n_cands: int = 5
    use_ste: bool = True
    sel_dropout: float = 0.0
    d_input: int = 512

    @classmethod
    def from_args(cls, args):
        return cls(n_layers=args.n_layers, n_heads=args.n_heads, d_model=args.d_model, d_model_ff=args.d_model_ff, enc_dropout=args.enc_dropout, use_text_query=args.use_text_query, use_text_cands=args.use_text_cands, n_cands=args.n_cands, use_ste=args.use_ste, sel_dropout=args.sel_dropout, d_input=args.d_input)

class ModalityEmbeddingsID(IntEnum):
    TEXT_QUESTION = 0
    TEXT_EMBEDDING = 1
    TEXT_UNUSED = 2
    VISUAL_EMBEDDING = 3
    VISUAL_UNUSED = 4

class ModalityEmbeddings(nn.Module):
    """
    Provides embeddings that indicate type of modality; for use with multimodal inputs for ATP. See atp.py for usage.
    """

    def __init__(self, d_model: int, use_text_query: bool=False, use_text_cands: bool=False, n_cands: int=5):
        """
        Details for each of these arguments are provided in ATPConfig.
        """
        super().__init__()
        self.d_model = d_model
        self.embedding = nn.Embedding(num_embeddings=len(ModalityEmbeddingsID), embedding_dim=d_model)
        self.use_text_query = use_text_query
        self.use_text_cands = use_text_cands
        self.n_cands = n_cands if use_text_cands else 0
        self.n_text_feats = 1 if use_text_query else 0
        if use_text_cands:
            self.n_text_feats += n_cands

    def forward(self, x: torch.tensor):
        """
        x: torch.tensor of size (L, N, D)
        returns modality embeddings for x of size (L, *, D)
        """
        L, N, D = x.size()
        n_frames = L - self.n_text_feats
        class_ids = []
        if self.use_text_query:
            class_ids = [ModalityEmbeddingsID.TEXT_QUESTION]
        if self.use_text_cands:
            class_ids.extend([ModalityEmbeddingsID.TEXT_EMBEDDING] * self.n_cands)
        class_ids.extend([ModalityEmbeddingsID.VISUAL_EMBEDDING] * n_frames)
        class_ids = torch.tensor(class_ids, dtype=torch.long, device=x.device).unsqueeze(-1)
        return self.embedding(class_ids)

class ATPEncoder(nn.Module):
    """
    The multimodal transformer encoder for the ATP model. For analysis purposes, the ATP encoder
    does not use any positional information (no positional encodings + transformer / self-attention)
    and is generally kept low-capacity. If the goal is raw accuracy (not analysis), you can relax these constraints.
    """

    def __init__(self, config: ATPConfig, **kwargs):
        """
        config: ATPConfig with parameters for the (transformer-based, atemporal) encoder for ATP.
        See ATPConfig documentation for details.
        """
        super().__init__()
        self.d_model = config.d_model
        self.dropout = nn.Dropout(p=config.enc_dropout)
        self.modality_encoding = ModalityEmbeddings(d_model=self.d_model, use_text_query=config.use_text_query, use_text_cands=config.use_text_cands, n_cands=config.n_cands)
        atp_encoder_layer = nn.TransformerEncoderLayer(d_model=self.d_model, nhead=config.n_heads, dim_feedforward=config.d_model_ff, dropout=config.enc_dropout, activation='relu')
        self.transformer_encoder = nn.TransformerEncoder(atp_encoder_layer, config.n_layers)

    def forward(self, x_inputs: torch.tensor):
        """
        x_inputs: torch.tensor of shape (L, N, D)
        """
        L, N, D = x_inputs.size()
        assert D == self.d_model, 'inputs dimension mismatch'
        x_encoded = x_inputs * math.sqrt(self.d_model)
        x_encoded += self.modality_encoding(x_encoded)
        x_encoded = self.dropout(x_encoded)
        x_encoded = self.transformer_encoder(x_encoded)
        return x_encoded


config = ATPConfig(
    n_layers=2,
    n_heads=2,
    d_model=128,
    d_model_ff=128,
    enc_dropout=0.1,
    use_text_query=False,
    use_text_cands=False,
    n_cands=5
)

INIT_ARGS = {
    'config': config
}

ARGS = {
    'x_inputs': torch.randn(7, 2, 128)
}

if __name__ == '__main__':
    encoder = ATPEncoder(**INIT_ARGS)
    output = encoder(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Expected shape: (7, 2, 128)")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ModalityEmbeddings Ran
ran on a synthesized input · constructed an object; no result was computed
stanfordvl/atp-video-language/src/atp.py
code served (permissive licence) · get_code("39bb83cfc59bca84")
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 an instance of its own class, ModalityEmbeddings: the run constructed the object and did not call it, so no result was computed. 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
from torch import nn

torch.manual_seed(0)

ARGS = {
    "d_model": 4,
    "use_text_query": True,
    "use_text_cands": True,
    "n_cands": 3,
}
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 ModalityEmbeddings — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ModalityEmbeddings 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
ATPSelectorModel Not yet run
unverified (harvested, no recorded run)
stanfordvl/atp-video-language/src/atp.py
code served (permissive licence) · get_code("b00fa686feea203d")
How it ran

no invocation recorded.

ModalityEmbeddingsID Not yet run
unverified (harvested, no recorded run)
stanfordvl/atp-video-language/src/atp.py
code served (permissive licence) · get_code("ada2c7f12b63bde0")
How it ran

no invocation recorded.

Also present in other papers' code

Of 5 samples lifted from this paper's linked repositories, 5 carry a function identity and 0 do not. Of the 5 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 5 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

What makes a video task uniquely suited for videos, beyond what can be understood from a single image? Building on recent progress in self-supervised image-language models, we revisit this question in the context of video and language tasks. We propose the atemporal probe (ATP), a new model for video-language analysis which provides a stronger bound on the baseline accuracy of multimodal models constrained by image-level understanding. By applying this model to standard discriminative video and language tasks, such as video question answering and text-tovideo retrieval, we characterize the limitations and potential of current video-language benchmarks. We find that understanding of event temporality is often not necessary to achieve strong or state-of-the-art performance, even compared with recent large-scale video-language models and in contexts intended to benchmark deeper video-level understanding. We also demonstrate how ATP can improve both video-language dataset and model design. We describe a technique for leveraging ATP to better disentangle dataset subsets with a higher concentration of temporally challenging data, improving benchmarking efficacy for causal and temporal understanding. Further, we show that effectively integrating ATP into full video-level temporal models can improve efficiency and state-of-the-art accuracy. 1

For agents

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

get_harvested_code_for_paper("2206.01720")
get_code_for_paper("2206.01720")
have("2206.01720")

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

curl https://syntology.ai/api/ran/2206.01720.json

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

[![Syntology run record](https://syntology.ai/api/ran/2206.01720.svg)](https://syntology.ai/paper/2206.01720)

Connect an agent — have() is free.