SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2409.19696 · NeurIPS · 2024

Vision-Language Models are Strong Noisy Label Detectors

Yu-Feng Li, Min-Ling Zhang, Jiang-Xin Shi, Tong Wei, Hao-Tian Li, Chun-Shu Li

arXiv · PDF · Open in the Atlas

Code that ran

2 of 8 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (1 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 1 has no contract verdict; 1 failure of our instrument; 6 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/2409.19696.json · badge: /api/ran/2409.19696.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
HotanLee/DeFT canonical (classifier label)
classifier basis: census
1 ran · 4 no recorded run
hotanlee/deft canonical (classifier label)
classifier basis: classifier_2025
1 ran · 2 no recorded run
FunctionStatusWhere it lives
CLIP_ViT Ran
ran on a synthesized input · constructed an object; no result was computed
hotanlee/deft/model/CLIP.py
code served (permissive licence) · get_code("579bce41ae2654f3")
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, CLIP_ViT: 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
import math

torch.manual_seed(0)

# Create a minimal mock CLIP model structure
class MockAttention(nn.Module):
    def __init__(self, embed_dim, num_heads):
        super().__init__()
        self.embed_dim = embed_dim
        self.num_heads = num_heads
        self.in_proj_weight = nn.Parameter(torch.randn(3 * embed_dim, embed_dim))
        self.in_proj_bias = nn.Parameter(torch.randn(3 * embed_dim))
        self.out_proj = nn.Linear(embed_dim, embed_dim)

class MockResBlock(nn.Module):
    def __init__(self, embed_dim, num_heads):
        super().__init__()
        self.ln_1 = nn.LayerNorm(embed_dim)
        self.attn = MockAttention(embed_dim, num_heads)
        self.ln_2 = nn.LayerNorm(embed_dim)
        self.mlp = nn.Sequential(
            nn.Linear(embed_dim, embed_dim * 4),
            nn.GELU(),
            nn.Linear(embed_dim * 4, embed_dim)
        )

class MockTransformer(nn.Module):
    def __init__(self, embed_dim, num_heads, num_layers):
        super().__init__()
        self.layers = num_layers
        self.resblocks = nn.ModuleList([
            MockResBlock(embed_dim, num_heads) for _ in range(num_layers)
        ])

class MockVisual(nn.Module):
    def __init__(self, embed_dim, num_heads, num_layers, patch_size, image_size):
        super().__init__()
        self.conv1 = nn.Conv2d(3, embed_dim, kernel_size=patch_size, stride=patch_size)
        num_patches = (image_size // patch_size) ** 2
        self.class_embedding = nn.Parameter(torch.randn(1, 1, embed_dim))
        self.positional_embedding = nn.Parameter(torch.randn(num_patches + 1, embed_dim))
        self.ln_pre = nn.LayerNorm(embed_dim)
        self.transformer = MockTransformer(embed_dim, num_heads, num_layers)
        self.ln_post = nn.LayerNorm(embed_dim)
        self.proj = nn.Parameter(torch.randn(embed_dim, embed_dim))

class MockCLIPModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.embed_dim = 4
        self.num_heads = 2
        self.num_layers = 2
        self.patch_size = 2
        self.image_size = 8
        self.dtype = torch.float32
        self.visual = MockVisual(
            embed_dim=self.embed_dim,
            num_heads=self.num_heads,
            num_layers=self.num_layers,
            patch_size=self.patch_size,
            image_size=self.image_size
        )

# Create mock clip_model
clip_model = MockCLIPModel()

# Create sample input tensor (batch_size=2, channels=3, height=8, width=8)
sample_input = torch.randn(2, 3, 8, 8)

ARGS = {
    "clip_model": clip_model,
}
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 CLIP_ViT — 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 CLIP_ViT 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
loss_coteaching Ran
ran; our contract draft was wrong, not the code
HotanLee/DeFT/main_small_loss.py
code served (permissive licence) · get_code("8da4e3661575688d")
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

torch.manual_seed(0)

# Create small tensors for the function
batch_size = 2
num_classes = 4

# loss: tensor of loss values (batch_size,)
loss = torch.randn(batch_size, requires_grad=True)

# y: predicted labels/logits (batch_size, num_classes)
y = torch.randn(batch_size, num_classes)

# t: target labels (batch_size,)
t = torch.randint(0, num_classes, (batch_size,))

# forget_rate: float between 0 and 1
forget_rate = 0.1

# ind: indices into the batch (batch_size,)
ind = torch.arange(batch_size)

# noise_or_not: boolean tensor indicating which samples are noisy (batch_size,)
# True = noisy, False = clean
noise_or_not = torch.tensor([False, True])

ARGS = {
    "loss": loss,
    "y": y,
    "t": t,
    "forget_rate": forget_rate,
    "ind": ind,
    "noise_or_not": noise_or_not,
}
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
Model Not yet run
unverified (harvested, no recorded run)
hotanlee/deft/model/CLIP.py
code served (permissive licence) · get_code("3c0179f2c08e061b")
How it ran

no invocation recorded.

ViT_Tuner Not yet run
unverified (harvested, no recorded run)
hotanlee/deft/model/CLIP.py
code served (permissive licence) · get_code("7d006c94f1996e46")
How it ran

no invocation recorded.

build_loader Not yet run
unverified (harvested, no recorded run)
HotanLee/DeFT/dataloader/dataloader_tiny_imagenet.py
code served (permissive licence) · get_code("15d083c0f312bf98")
How it ran

no invocation recorded.

eval_train Not yet run
unverified (harvested, no recorded run)
HotanLee/DeFT/main_real_baseline.py
code served (permissive licence) · get_code("514531c5f0f598b8")
How it ran

no invocation recorded.

test Not yet run
unverified (harvested, no recorded run)
HotanLee/DeFT/main_small_loss.py
code served (permissive licence) · get_code("db80cb31193f92b1")
How it ran

no invocation recorded.

test Not yet run
unverified (harvested, no recorded run)
HotanLee/DeFT/main_phase1.py
code served (permissive licence) · get_code("c8a8e3f0b597c1e7")
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, 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

Recent research on fine-tuning vision-language models has demonstrated impressive performance in various downstream tasks. However, the challenge of obtaining accurately labeled data in real-world applications poses a significant obstacle during the fine-tuning process. To address this challenge, this paper presents a Denoising Fine-Tuning framework, called DEFT, for adapting vision-language models. DEFT utilizes the robust alignment of textual and visual features pretrained on millions of auxiliary image-text pairs to sieve out noisy labels. The proposed framework establishes a noisy label detector by learning positive and negative textual prompts for each class. The positive prompt seeks to reveal distinctive features of the class, while the negative prompt serves as a learnable threshold for separating clean and noisy samples. We employ parameter-efficient fine-tuning for the adaptation of a pre-trained visual encoder to promote its alignment with the learned textual prompts. As a general framework, DEFT can seamlessly fine-tune many pre-trained models to downstream tasks by utilizing carefully selected clean samples. Experimental results on seven synthetic and real-world noisy datasets validate the effectiveness of DEFT in both noisy label detection and image classification tasks. Our source code is available at https://github.com/HotanLee/DeFT.

For agents

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

get_harvested_code_for_paper("2409.19696")
get_code_for_paper("2409.19696")
have("2409.19696")

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

curl https://syntology.ai/api/ran/2409.19696.json

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

[![Syntology run record](https://syntology.ai/api/ran/2409.19696.svg)](https://syntology.ai/paper/2409.19696)

Connect an agent — have() is free.