Pith. sign in

REVIEW 4 major objections 5 minor 27 references

Embedding-based Retrieval in Multimodal Content Moderation

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper claims that an embedding-based retrieval system can catch emerging harmful video trends more accurately and at far lower cost than repeatedly retraining a classifier.

desk verdict Useful industry EBR-for-moderation report, but Table 1's near-perfect numbers may be inflated by seed leakage into the evaluation set; needs a cleanup before the claims can be trusted. read the letter →

arxiv 2507.01066 v1 pith:IQLU3URR submitted 2025-06-30 cs.IR cs.CVcs.LG

Hanzhong Liang , Jinghao Shi , Xiang Shen , Zixuan Wang , Vera Wen , Ardalan Mehrani , Zhiqian Chen , Yifan Wu
show 1 more author
Zhixin Zhang
This is my paper
classification cs.IRcs.CVcs.LG
keywords Embedding-basedretrievalContentmoderationSupervisedcontrastivelearningMultimodalembeddingVideoTrenddetectionSimilaritysearchOnlinedeployment
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that for content moderation on short-video platforms, retrieval by embedding similarity can outperform classification when the task is to catch a new harmful trend quickly. Its system finds a few known-harmful 'seed' videos in the emerging trend, embeds them and all new videos, and scores each new video by its highest cosine similarity to a seed; high similarity triggers moderation actions. On 25 real-world trends, the reported multimodal EBR reaches ROC-AUC 0.996 and PR-AUC 0.955 versus the classification baseline's 0.853 and 0.350, and online deployment raised total action volume by 10.32% while cutting trend-handling cost by over 80%. The authors claim this works because supervised contrastive learning organizes the embedding space around risk semantics rather than raw visual similarity, so a few seeds capture the trend without retraining the model.

What carries the argument

The load-bearing mechanism is Supervised Contrastive Learning (SCL), a training objective that pulls together videos sharing the same risk label and pushes apart videos with different labels, so the embedding space is organized by risk semantics rather than only by visual similarity. The retrieval system built on it has five components: seed selection (density-based DBSCAN clustering for centroid-proximity seeds, historical-precision filtering, and moderator-supplied golden seeds), the embedding model (a ViT for single-modal, or ViT plus a RoBERTa text encoder with cross-attention for multimodal), a retrieval service that computes pairwise cosine similarity between seeds and candidate videos, an auto-action service that thresholds the maximum similarity score to decide flagging, restriction, or escalation, and a feedback loop that monitors Top-K precision and adjusts seeds and thresholds in real time.

What would settle it

Take the same 25 evaluation trends, give the classifier model the same 5% seed videos per trend as a few-shot prompt or a quick fine-tune, and compare ROC-AUC and PR-AUC against the EBR numbers; if the classifier closes the gap, the retrieval advantage is mainly seed-information rather than the retrieval mechanism itself. Alternatively, measure EBR on a trend for which no seeds exist yet, to test whether the system can act before humans identify the trend.

Watch

Extended reading notes

Core claim

The central claim is that a similarity-retrieval system, scoring each new video by its maximum cosine similarity to a small set of seed videos, detects unseen harmful trends more accurately than a trained multimodal classifier and requires no model retraining. The offline evaluation compares a 25-trend dataset unseen to both systems: for each trend, the EBR system is given 5% of the trend's videos as seeds and retrieves the top 200 candidates; the multimodal EBR model scores ROC-AUC 0.996 and PR-AUC 0.955, against the classifier's 0.853 and 0.350. In production the same system brought total moderation actions up 10.32% and reduced the cost of trend handling by over 80%, and on three manually reviewed trends it recalled over 95% of the harmful videos while the classification-based model recalled under 3%. The paper concludes that EBR works as a hot-fix complement to classifiers, adapting to new trends within a day rather than the roughly five days a classifier update takes.

Load-bearing premise

The load-bearing premise is that the comparison is fair: the retrieval system gets 5% of each trend's videos as seeds while the baseline classifier gets no examples from the evaluation trends; if the classifier were given the same few seed examples, the size of the reported gap could change.

Editorial extensions

If this is right

  • EBR can be deployed as a hot-fix layer alongside a classification model, catching a new harmful trend within one day instead of the average five days a classifier update takes.
  • A seed pool of 5% to 10% of a trend's videos is enough to represent it; adding more seeds beyond 10% gives only marginal gains.
  • Multimodal embeddings beat single-modal ones for these risk trends, so incorporating the video's textual signals improves retrieval accuracy.
  • Because the retrieval score is tied to concrete seed examples, the system offers reviewers interpretable evidence (the retrieved seeds) rather than an opaque class score.
  • The similarity threshold can be set per action type, so the same system handles flagging for review, content restriction, and escalation with one embedding space.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The reported offline comparison is asymmetric: EBR sees 5% of each evaluation trend's videos as seeds while the classification baseline has no examples from those trends; a few-shot-adapted classifier might narrow the gap, so the headline superiority should be read as retrieval-with-seeds versus a fixed classifier.
  • The near-ceiling offline numbers suggest the evaluation trends are semantically coherent clusters in the embedding space; on broader, more heterogeneous risk categories the retrieval advantage may be smaller.
  • A cold-start limitation is implicit: if no one has identified seeds for a brand-new trend, EBR cannot start retrieving, so it complements rather than replaces classification for true zero-day detection.
  • The 'over 80% cost reduction' counts avoided classifier retraining but not the cost of embedding inference at platform scale or the human time spent curating moderator seeds; a full cost model would include both.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes an Embedding-Based Retrieval (EBR) system for multimodal content moderation, in which supervised contrastive learning trains single- and multi-modal embedding models, and new videos are scored by their maximum cosine similarity to a small set of seed videos from an emerging trend. The authors report offline experiments on 25 trends showing EBR achieving ROC-AUC 0.996 and PR-AUC 0.955 versus 0.853 and 0.350 for a multimodal classification baseline (Table 1), plus online deployment results claiming a 10.32% increase in total action volume and over 80% cost reduction (Sections 4.4.1 and 4.4.2). The central claim is that seed-based retrieval can complement classification for fast, low-cost adaptation to emerging harmful trends.

Significance. If the evaluation is sound, this is a practically valuable contribution: it is, to the authors' knowledge, the first deployment-scale EBR system for content moderation, it is validated on production data, and the contrastive-learning comparison against CLIP and MoCo (Table 2) is informative. The online results, including a 10.32% action-volume increase and an 80% cost reduction, suggest real operational impact. However, the headline offline numbers are currently not trustworthy because the evaluation protocol as written allows seed leakage and compares EBR against a baseline that lacks the same label access. The paper's significance therefore hinges on correcting the offline evaluation and adding a like-for-like baseline; with those fixes, the contribution would be solid for an industry-track venue. The use of deployed, production-scale evidence and the explicit reporting of cost and action metrics are strengths that should be retained.

major comments (4)
  1. [Section 4.2.1, Section 4.2.2, Table 1] The evaluation protocol does not state that the 5% per-trend seed videos are removed from the candidate/evaluation set. Since the paper defines each video's EBR score as its maximum cosine similarity among all seed videos, any seed remaining in the evaluation set is a trivial positive with similarity 1.0 to itself. This would guarantee perfect recall on at least 5% of positives and would inflate ROC-AUC, PR-AUC, F1, and P@200 without reflecting generalization to unseen trend content. The statement that the evaluation dataset is 'unseen to both the EBR and online classification model' is ambiguous because EBR is explicitly given seed videos from that dataset at inference time. This issue must be resolved by stating clearly that seeds are excluded from the evaluated candidate set, and all Table 1 numbers must be recomputed under that protocol.
  2. [Section 4.2.1, Table 1] The comparison is not apples-to-apples: EBR receives 5% of each trend's positive videos as seeds, while the baseline is a frozen multimodal classifier with no access to any evaluation-trend examples. The claim that EBR 'significantly outperforms' the baseline therefore conflates seed-label access with retrieval quality. The authors should add a control in which the baseline is given the same seed information, for example through few-shot adaptation or retrieval-augmented classification, or evaluate EBR in a zero-shot mode without trend-specific seeds. Without such a control, the magnitude of the reported gap (PR-AUC 0.350 to 0.955) is not interpretable as a superiority of the retrieval mechanism.
  3. [Table 1, Section 4.2.3] The paper aggregates results over 25 trends whose sizes range from 200 to 20K videos and whose positive-to-negative ratios vary from 1:50 to 1:1, but it reports only aggregate metrics with no per-trend breakdown, error bars, or significance tests. Because the leakage issue could affect trends of different sizes very differently, the aggregate near-perfect numbers cannot be assessed without per-trend results. The authors should report per-trend ROC-AUC, PR-AUC, F1, and P@200, along with confidence intervals or a variance measure.
  4. [Section 4.4.1, Table 3] The online trend comparison in Table 3 reports EBR recall above 95% versus model recall below 3%, but the denominator of these recall numbers and the overlap between the EBR seed set and the counted 'total volume' are not defined. If the seed videos used by EBR were part of the manually evaluated trend volume, the recall figure would again include trivially retrieved seeds. The authors should specify how the 'total volume' was determined, whether seed videos were excluded from the manual review set, and how the classification-model recall was computed, so that the online comparison is not subject to the same leakage concern.
minor comments (5)
  1. [Section 4.2.1] There is a typo: 'comparion' should be 'comparison'. The phrase 'every seed recall top 200 candidates' should be clarified, for example as 'for every seed, retrieve the top 200 candidates, with duplicate videos across seeds removed'.
  2. [Section 4.4.2] The paper calls EBR 'train-free', but Section 4.1 describes training the embedding models on 430M videos for 20 days. It would be clearer to say that EBR is 'train-free at trend-handling time' or 'requires no per-trend model retraining'.
  3. [Figure 2] Figure 2 shows the effect of seed percentage on PR-AUC and F1, but the axes are not fully described and no error bars or per-seed repetition details are given. Adding these would make the '5%-10% seed range' recommendation more reproducible.
  4. [Section 3.1.2] The description of the multimodal architecture is brief: it would help to state how video frames are sampled and aggregated, how the cross-attention module is structured, and how the text encoder output is fused with the vision output before projection.
  5. [Section 4.3.1] Table 2 compares SCL with CLIP, MoCo, and a ResNet-based cross-entropy model, but the 'same dataset' is not otherwise characterized; specifying the training data, label granularity, and evaluation tasks would strengthen the ablation.

Circularity Check

1 steps flagged · score 6.0 of 10

Table 1's EBR superiority is partially circular: the 5% per-trend seeds are drawn from the same evaluation set and each video's score is the max cosine similarity to those seeds, so every seed trivially self-scores 1.0.

  1. fitted input called prediction [Section 4.2.1, Section 4.2.2, Table 1]
    "All the evaluation dataset is unseen to both the EBR and online classification model. For EBR, we randomly select 5% of videos per trend as seed samples. And every seed recall top 200 candidates. ... For EBR evaluation, each video's classification score is its maximum cosine similarity among all seed videos."

    The 5% seed videos are drawn from the evaluation trends, and the EBR score of any video is defined as its maximum cosine similarity to all seed videos. A seed video compared with itself has cosine similarity exactly 1.0, so by construction every seed in the evaluation set receives a perfect score and is guaranteed to be retrieved. Since the paper does not state that the seed videos are removed from the candidate/evaluation set, 5% of positive examples trivially attain score 1.0, inflating the ROC-AUC (0.996), PR-AUC (0.955), F1 (0.945), and P@200 (0.835) reported in Table 1.

full rationale

The paper contains no analytic derivation from fitted constants; it is an empirical systems paper, so the main circularity risk is in the offline evaluation. The single concrete circular step is the seed/evaluation overlap described above: the EBR score function is built from the same 5% of trend videos that are also in the evaluation pool, making the seed videos' scores 1.0 by definition. This directly affects the central claim of Table 1. The only self-citation ([20], CPFD) is background on video classification and is not load-bearing. The SCL loss (Eq. 1) is a standard supervised contrastive objective and is not circular. The offline comparison's asymmetry (EBR receives trend seeds while the baseline classifier receives none) is an evaluation-design concern rather than circularity per se, but it compounds the leakage problem. Overall, the reported near-perfect retrieval numbers are partially reduced to self-retrieval, so the central empirical claim is not fully self-contained.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central claim is empirical and system-level. It depends on domain assumptions about how well risk-title supervision transfers to unseen trends, how representative the 25-trend evaluation set is, and that the evaluation set is truly disjoint from the 430M-video training set. All quantitative results rely on proprietary data that has not been released, so these assumptions cannot be independently verified.

free parameters (5)
  • Seed ratio for offline evaluation = 5% of videos per trend
    The offline evaluation selects 5% of videos per trend as seeds (Section 4.2.1); the ablation (Section 4.3.2) shows performance saturates around 10%, indicating a hand-chosen operating point.
  • Retrieval top-K = 200 candidates per seed
    The system retrieves top 200 candidates per seed (Section 4.2.1); P@200 is used as an evaluation metric, so the truncation is built into the reported numbers.
  • Auto-action similarity thresholds = unspecified
    The Auto-Action service triggers actions based on predefined similarity thresholds (Section 3.2); the threshold values are not reported and would be tuned operationally.
  • Historical seed precision threshold = unspecified
    Historical Data Seed Selection declares a seed 'good' if its past precision exceeds a threshold (Section 3.2.1); the threshold is not specified.
  • SupCon temperature tau = unspecified
    The supervised contrastive loss (Eq. 1) includes a temperature parameter tau; its value and scheduling are not reported.
assumptions (6)
  • domain assumption Videos sharing a risk title should have similar embeddings
    The SCL training (Eq. 1) defines positive pairs by shared risk labels, assuming risk-title semantics align with visual/multimodal similarity in embedding space (Section 3.1.1).
  • domain assumption Similarity to a small set of seed videos generalizes to emerging harmful trends
    The entire EBR mechanism assumes the embedding space clusters novel harmful content near the seed videos, so max-cosine-similarity scoring transfers to unseen trends (Section 3.2).
  • domain assumption The 25 evaluation trends and random negatives represent real emerging harmful content
    The offline metrics aggregate 25 trends of size 200 to 20K with positive-to-negative ratios from 1:50 to 1:1 (Section 4.2.1); no details on trend selection or labeling are given, so representativeness is assumed.
  • domain assumption The evaluation dataset is truly unseen by both the embedding model and the classifier
    The paper states this (Section 4.2.1) but provides no data release or algorithmic evidence to rule out overlap with the 430M video training set, so the claim is unverifiable.
  • domain assumption Past precision approximates future precision in seed selection
    Historical Data Seed Selection assumes p_{t+1} approx p_{[t-x,t]} (Eq. 3); this stationarity assumption is load-bearing for the feedback loop but is not validated.
  • standard math Standard contrastive learning loss and architectures are valid
    The method builds on SupCon (Khosla et al. 2020), ViT, RoBERTa, and cross-attention modules from prior literature; these are standard background.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Embedding-based Retrieval in Multimodal Content Moderation." pith.science (2026). https://pith.science/paper/IQLU3URR

@misc{pith2026250701066,
  author       = {Pith},
  title        = {Pith review of: Embedding-based Retrieval in Multimodal Content Moderation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IQLU3URR}},
  note         = {Machine review of arXiv:2507.01066}
}
read the original abstract

Video understanding plays a fundamental role for content moderation on short video platforms, enabling the detection of inappropriate content. While classification remains the dominant approach for content moderation, it often struggles in scenarios requiring rapid and cost-efficient responses, such as trend adaptation and urgent escalations. To address this issue, we introduce an Embedding-Based Retrieval (EBR) method designed to complement traditional classification approaches. We first leverage a Supervised Contrastive Learning (SCL) framework to train a suite of foundation embedding models, including both single-modal and multi-modal architectures. Our models demonstrate superior performance over established contrastive learning methods such as CLIP and MoCo. Building on these embedding models, we design and implement the embedding-based retrieval system that integrates embedding generation and video retrieval to enable efficient and effective trend handling. Comprehensive offline experiments on 25 diverse emerging trends show that EBR improves ROC-AUC from 0.85 to 0.99 and PR-AUC from 0.35 to 0.95. Further online experiments reveal that EBR increases action rates by 10.32% and reduces operational costs by over 80%, while also enhancing interpretability and flexibility compared to classification-based solutions.

Figures

Figures reproduced from arXiv: 2507.01066 by the authors.

Figure 1
Figure 1. EBR System Design. When a trends emerge, the Seed Selection module selects seed videos, which are then processed by [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. EBR System with different seed percentage. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

27 extracted references · 13 canonical work pages

  1. [1]

    Le Binh, Rajat Tandon, Chingis Oinar, Jeffrey Liu, Uma Durairaj, Jiani Guo, Spencer Zahabizadeh, Sanjana Ilango, Jeremy Tang, Fred Morstatter, et al. 2022. Samba: Identifying Inappropriate Videos for Young Children on YouTube. In Proceedings of the 31st ACM International Conference on Information & Knowledge Management. 88–97

    work page 2022

  2. [2]

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A simple framework for contrastive learning of visual representations. In Interna- tional conference on machine learning . PMLR, 1597–1607

    2020

  3. [3]

    Mithun Das, Rohit Raj, Punyajoy Saha, Binny Mathew, Manish Gupta, and Ani- mesh Mukherjee. 2023. Hatemm: A multi-modal dataset for hate video classi- fication. In Proceedings of the International AAAI Conference on Web and Social Media, Vol. 17. 1014–1023

    work page 2023

  4. [4]

    Alexey Dosovitskiy. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929 (2020)

    arXiv 2020

  5. [5]

    Martin Ester, Hans-Peter Kriegel, Jörg Sander, Xiaowei Xu, et al. 1996. A density- based algorithm for discovering clusters in large spatial databases with noise. In kdd, Vol. 96. 226–231

    1996

  6. [6]

    Vaishali U Gongane, Mousami V Munot, and Alwin D Anuse. 2022. Detection and moderation of detrimental content on social media platforms: current status and future directions. Social Network Analysis and Mining 12, 1 (2022), 129

    work page 2022

  7. [7]

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. 2020. Mo- mentum contrast for unsupervised visual representation learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 9729–9738

    2020

  8. [8]

    Jui-Ting Huang, Ashish Sharma, Shuying Sun, Li Xia, David Zhang, Philip Pronin, Janani Padmanabhan, Giuseppe Ottaviano, and Linjun Yang. 2020. Embedding- based retrieval in facebook search. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining . 2553–2561

    2020

Show all 27 references
  1. [9]

    Akshay Kekuda, Yuyang Zhang, and Arun Udayashankar. 2024. Embedding based retrieval for long tail search queries in ecommerce. In Proceedings of the 18th ACM Conference on Recommender Systems . 771–774

    2024

  2. [10]

    Prannay Khosla, Piotr Teterwak, Chen Wang, Aaron Sarna, Yonglong Tian, Phillip Isola, Aaron Maschinot, Ce Liu, and Dilip Krishnan. 2020. Supervised contrastive learning. Advances in neural information processing systems 33 (2020), 18661– 18673

    2020

  3. [11]

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. 2022. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. In International conference on machine learning . PMLR, 12888–12900

    2022

  4. [12]

    Sen Li, Fuyu Lv, Taiwei Jin, Guli Lin, Keping Yang, Xiaoyi Zeng, Xiao-Ming Wu, and Qianli Ma. 2021. Embedding-based product retrieval in taobao search. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. 3181–3189

    2021

  5. [13]

    Juexin Lin, Sachin Yadav, Feng Liu, Nicholas Rossi, Praveen R Suram, Satya Chem- bolu, Prijith Chandran, Hrushikesh Mohapatra, Tony Lee, Alessandro Magnani, et al. 2024. Enhancing Relevance of Embedding-based Retrieval at Walmart. In Proceedings of the 33rd ACM International C...

    2024

  6. [14]

    Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692 (2019)

    2019 arXiv

  7. [15]

    Huan Ma, Changqing Zhang, Huazhu Fu, Peilin Zhao, and Bingzhe Wu. 2023. Adapting large language models for content moderation: Pitfalls in data engi- neering and supervised fine-tuning. arXiv preprint arXiv:2310.03400 (2023)

    2023 arXiv

  8. [16]

    Sankha Subhra Mullick, Mohan Bhambhani, Suhit Sinha, Akshat Mathur, Somya Gupta, and Jidnya Shah. 2023. Content moderation for evolving policies using binary question answering. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume ...

    2023

  9. [17]

    notAI.tech. 2024. NudeNet: lightweight Nudity detection. https://github.com/ notAI-tech/NudeNet

    2024

  10. [18]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learni...

    2021

  11. [19]

    Jiahui Shi, Vivek Chaurasiya, Yozen Liu, Shubham Vij, Yan Wu, Satya Kanduri, Neil Shah, Peicheng Yu, Nik Srivastava, Lei Shi, et al. 2023. Embedding Based Retrieval in Friend Recommendation. In Proceedings of the 46th International ACM SIGIR Conference on Research and Developm...

    2023

  12. [20]

    Jinghao Shi, Xiang Shen, Kaili Zhao, Xuedong Wang, Vera Wen, Zixuan Wang, Yifan Wu, and Zhixin Zhang. 2024. CPFD: Confidence-aware Privileged Fea- ture Distillation for Short Video Classification. In Proceedings of the 33rd ACM International Conference on Information and Knowl...

    2024

  13. [21]

    Mengyang Wu, Yuzhi Zhao, Jialun Cao, Mingjie Xu, Zhongming Jiang, Xuehui Wang, Qinbin Li, Guangneng Hu, Shengchao Qin, and Chi-Wing Fu. 2024. ICM- Assistant: Instruction-tuning Multimodal Large Language Models for Rule-based Explainable Image Content Moderation. arXiv preprint...

    2024 arXiv

  14. [22]

    Hu Xu, Gargi Ghosh, Po-Yao Huang, Dmytro Okhonko, Armen Aghajanyan, Florian Metze, Luke Zettlemoyer, and Christoph Feichtenhofer. 2021. VideoCLIP: Contrastive Pre-training for Zero-shot Video-Text Understanding. In Proceedings of the 2021 Conference on Empirical Methods in Nat...

    2021 doi

  15. [23]

    Meng Ye, Karan Sikka, Katherine Atwell, Sabit Hassan, Ajay Divakaran, and Malihe Alikhani. 2023. Multilingual Content Moderation: A Case Study on Reddit. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, Andreas Vla...

    2023

  16. [24]

    Jialin Yuan, Ye Yu, Gaurav Mittal, Matthew Hall, Sandra Sajeev, and Mei Chen

  17. [25]

    Yan Zeng, Xinsong Zhang, and Hang Li. 2022. Multi-Grained Vision Language Pre-Training: Aligning Texts with Visual Concepts. InInternational Conference on Machine Learning. PMLR, 25994–26009

    2022

  18. [26]

    Yuhui Zhang, Alyssa Unell, Xiaohan Wang, Dhruba Ghosh, Yuchang Su, Ludwig Schmidt, and Serena Yeung-Levy. 2024. Why are Visually-Grounded Language Models Bad at Image Classification? Conference on Neural Information Processing Systems (NeurIPS) (2024)

    2024

  19. [2024]

    In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision

    Rethinking Multimodal Content Moderation from an Asymmetric An- gle with Mixed-modality. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. 8532–8542

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.