Pith. sign in

REVIEW 3 major objections 4 minor 43 references

Seeing the Abstract: Translating the Abstract Language for Vision Language Models

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read CLIP-style models under-represent abstract fashion language, and a training-free PCA shift can close most of that gap, beating fine-tuned models on retrieval.

desk verdict A real and measurable abstract-language bias in VLM text encoders, with a mostly working training-free corrective shift that needs one equation clarified and one overclaim reined in. read the letter →

arxiv 2505.03242 v1 pith:A4ETHOAO submitted 2025-05-06 cs.CV cs.AI

classification cs.CVcs.AI
keywords visionlanguagemodelsabstracttext-to-imageretrievalfashionrepresentationshifttraining-freeadaptationLLMrewritingPCAtranslation
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

Fashion descriptions are rich in abstract adjectives—words like 'sexy', 'airy', and 'chic'—that occur about as often as concrete ones, carry information concrete attributes do not, and make text queries more discriminative. The paper argues that CLIP-style vision-language models under-represent this abstract language because their pre-training text is concrete-heavy, and that fine-tuning does not fully repair the gap. It proposes ACT, a training-free pipeline that has an LLM rewrite an abstract query toward concrete wording and then adds a PCA-derived shift in the VLM's text-embedding space, moving the query toward the embeddings of concrete captions. On text-to-image retrieval in the fashion domain, ACT beats zero-shot and fine-tuned baselines in both same-dataset and cross-dataset settings. If right, this means much of the abstract-language gap in such models is a systematic, low-dimensional embedding shift that can be corrected without retraining.

What carries the argument

The load-bearing mechanism is the Abstract-to-Concrete Translator (ACT), a two-phase, training-free procedure. In preparation, paired embeddings $h^C_s$ and $h^A_s$ are obtained from captioning-model captions and original abstract descriptions of the same items; their standardized difference, $\Delta_{\text{A-C}} = H_C - H_A$, is projected with Principal Component Analysis onto its top $k = 600$ directions, giving a shift projector $W$ along with mean and standard-deviation statistics. At inference, the query is first rewritten by a frozen LLM, embedded, then shifted by $\hat{h}^C_q = h_{q'} + (N(h_{q'}) W W^\top \sigma_\Delta + \mu_\Delta)$. This addition is what carries the argument: a single linear correction, learned once per dataset and VLM, moves abstract query embeddings toward the concrete caption region of the latent space, and cosine retrieval with the shifted embedding outperforms fine-tuned models.

What would settle it

Build a control set in which the captioning-model captions are length- and style-matched to the original descriptions, for example by truncating or padding them, and recompute the PCA shift. If the retrieval gain disappears, the shift encodes captioning style rather than abstractness. As a second check, apply the same shift to the LLM-rewritten, already-concrete queries: if it lifts retrieval as much as it lifts abstract queries, the shift is not specifically translating abstract meaning.

Watch

Extended reading notes

Core claim

The central claim is that current VLMs, both general-purpose and fashion-specific, systematically under-represent abstract-oriented language, and that the gap can be largely closed without weight updates. The paper builds an Abstract-Concrete database by captioning fashion images with a frozen captioning model, then computes the embedding difference between the captioning-model captions and the original abstract descriptions for the same images. The dominant directions of this difference, extracted with PCA, define a shift that is added to the query representation at inference, after an LLM has already rewritten the query in more concrete terms. On DeepFashion retrieval, ACT-df reaches H@1 of 0.437, above the best fine-tuned model (SigLIP-ft-df at 0.417) and far above zero-shot SigLIP at 0.311; in the cross-dataset setting, ACT-facad reaches 0.428, above every fine-tuned baseline. The paper also claims the improvement is consistent across model families and scales, with an average +4.9% H@1 gain.

Load-bearing premise

The method assumes that the difference between captioning-model captions and original abstract descriptions is dominated by the abstract-to-concrete semantic shift, and that one linear direction learned from that difference transfers to new queries, rather than reflecting the captioning model's shorter length, repetitive phrasing, or narrower vocabulary.

Editorial extensions

If this is right

  • Same-dataset retrieval on DeepFashion would rise from 0.311 to 0.437 H@1 with SigLIP, beating the best fine-tuned model (0.417) with no weight updates.
  • Cross-dataset use, building the shift on FACAD and testing on DeepFashion, would reach 0.428 H@1, nearly matching the same-dataset version and beating all fine-tuned baselines.
  • The gain would generalize across model families and scales, with an average +4.9% H@1 over SigLIP, CLIP, O-CLIP and EVA-CLIP backbones.
  • Both pipeline components would be necessary: LLM rewriting alone adds about +10% H@1 and the representation shift alone about +3.7% H@1, with the best result when combined.
  • Fashion text-to-image search would work with natural, abstract user phrasing rather than requiring users to supply concrete attribute lists.

Reading between the lines

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

  • If the shift is a genuine semantic correction, the same preparation could be run for other under-represented registers—metaphor, emotive, subculture-specific—turning ACT into a generic 'domain dialect' adapter for CLIP-like encoders.
  • The dependence on a captioning model means ACT's notion of 'concrete' is whatever that model's vocabulary captures; using human-written concrete paraphrases or a captioner with controlled length/style would test how much of the gain is style matching.
  • The supplementary's FACAD failure suggests the LLM rewriting step, not the PCA shift, is the bottleneck for terse, ungrammatical queries; a rewriter trained for concise fashion text would likely extend ACT's gains there.
  • Since the shift is computed once per VLM and dataset and applied as an embedding addition, it could be shipped as a lightweight per-domain plug-in, potentially transferring to composed retrieval or text-to-video search without fine-tuning.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper studies abstract-oriented language in vision-language models, focusing on the fashion domain. It presents a statistical analysis of adjectives in DeepFashion, FACAD, and LAION-400M, arguing that abstract adjectives are frequent in fashion descriptions, carry information not captured by concrete attributes, and are useful for text-to-image retrieval. It then proposes ACT (Abstract-to-Concrete Translator), a training-free method that (i) builds an Abstract-Concrete database by captioning fashion images with a frozen VLM, (ii) fits a PCA-based shift between text embeddings of original abstract descriptions and generated concrete captions, and (iii) at inference rewrites the user query with an LLM, embeds it, and applies the learned shift. The paper reports that ACT outperforms zero-shot and fine-tuned VLMs in same-dataset and cross-dataset settings on DeepFashion, and shows consistent gains across multiple VLM families and scales. The main empirical claim is ACT-df reaching H@1 = 0.437 on DeepFashion, above SigLIP-ft-df at 0.417, and ACT-facad reaching H@1 = 0.428 in the FACAD-to-DeepFashion setting.

Significance. If the central claim is validated, the paper is a useful and interesting contribution: it provides a large-scale empirical analysis of abstract versus concrete language in fashion VLMs and demonstrates a plug-and-play, training-free alternative to fine-tuning for closing part of the abstract-language performance gap. The paper ships code and evaluates across many backbones (SigLIP, O-CLIP, EVA-CLIP) and model sizes, which makes the empirical basis unusually broad. The finding that a linear PCA shift in text-embedding space, combined with LLM rewriting, can transfer across datasets is a genuinely useful result. However, the significance is contingent on resolving the reproducibility issue in Eq. (8) and on a more balanced statement of the cross-dataset claims, since the supplementary results in the opposite direction (DeepFashion-to-FACAD) do not show the same superiority.

major comments (3)
  1. [Sec. 4, Eq. (8)] Equation (8) is not reproducible as written. The standardization N(h_q') is described as using 'the mean and standard deviation among query representations,' but the paper never states where those statistics come from. If they are computed on the evaluation query set, then ACT is transductive: the embedding of each test query is centered and scaled using information from all other test queries, a form of test-set information that is not available to the zero-shot or fine-tuned baselines. This issue is load-bearing: Table 3 shows ACT-df exceeds SigLIP-ft-df by H@1 = 0.437 versus 0.417, while Table 5 shows that the language-rewriting component alone reaches only H@1 = 0.411. The advantage therefore comes from the representation-shift term, which is exactly the potentially transductive part. The authors must specify whether the query mean and standard deviation are computed on the training split, on the A-C database, or on the evaluation query set, and they should report results with query statistics restricted to the training split only.
  2. [Supp. C.2 and Abstract] The abstract and Section 5 claim that ACT 'outperforms the fine-tuned VLMs in both same- and cross-dataset settings,' but the DeepFashion-to-FACAD cross-dataset results in Supp. Table 3 contradict this: ACT-df achieves H@1 = 0.426 and ACT-df without language rewriting achieves H@1 = 0.460, both below SigLIP-ft-df at H@1 = 0.482. This is not a minor gap; it is a direct counterexample to the unqualified claim and also shows that the LLM-rewriting component can substantially hurt in this direction. The paper should either restrict the cross-dataset claim to the FACAD-to-DeepFashion direction, or add a balanced analysis of the asymmetry and explain under what conditions ACT can be expected to help.
  3. [Sec. 4, Eqs. (4)-(6), and Table 5] The central interpretation that the PCA shift W captures an abstract-to-concrete semantic shift is not yet established, because the paired differences h_C - h_A also encode the captioning model's shorter length, simpler syntax, restricted vocabulary, and the LLM's stylistic rewriting. The paper does not provide a control that separates abstractness from these confounds, so the retrieval gains could be explained by distribution matching toward caption-style text rather than by translating abstract meaning. A concrete diagnostic would be to apply the shift to queries that are already fully concrete (or whose LLM rewriting contains no abstract adjectives); if W is a genuine abstract-to-concrete map, the shift should be approximately inert on such queries. At minimum, the paper should report length and lexical statistics for original, rewritten, and captioned text, and show that the selected PCA directions are tied to abstract-concrete contrasts rather than to sentence length or caption style.
minor comments (4)
  1. [Eq. (8)] The expression 'W W^T * sigma_Delta + mu_Delta' mixes matrix and vector notation; please specify whether sigma_Delta is a scalar, a per-component vector, or a diagonal matrix, and how the addition broadcasts. This is needed for exact reproducibility.
  2. [Supp. B.2] The manual reclassification of context-dependent adjectives such as 'cool' and 'sharp' is a reasonable domain adaptation, but the number of reclassified words and a robustness check without the manual edits would help assess how much the abstract/concrete statistics depend on this choice.
  3. [Sec. 5, Table 3] The main comparisons are reported without error bars or a significance test; the headline advantage over SigLIP-ft-df is 0.020 in H@1, so a confidence interval, per-seed variance, or a paired test would substantially strengthen the claim that ACT 'outperforms' the best fine-tuned model.
  4. [Sec. 5.1 and Supp. C.2] The DeepFashion-to-FACAD results and the associated failure analysis of language rewriting should be moved from the supplementary into the main text, because they materially qualify the cross-dataset claim made in the abstract and in Section 5.1.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ACT's PCA shift is fit on training data and tested on held-out queries.

full rationale

The derivation is self-contained and the reported gains are not forced by construction. Abstractness is defined by an external concreteness lexicon (Brysbaert et al. [5]); ACT's shift W is fit on a training A-C database built from captioning-model outputs (Eqs. 2-6) and is then applied to held-out queries (Eq. 8). The retrieval evaluation uses disjoint train/test splits: ACT-df uses DeepFashion train for W and is tested on the DeepFashion eval set; ACT-facad uses FACAD and is tested on DeepFashion. The ablation (Tab. 5) quantifies each component's marginal contribution, and Supp. Sec. C.2 reports a cross-dataset case (DeepFashion to FACAD) where language rewriting alone hurts, showing the method is not guaranteed to win by the construction of its inputs. The only caveat is Eq. (8)'s N(·), whose query mean/std are not specified: if they are computed on the evaluation query set, the method would be transductive, but that is a reproducibility/validity question, not a circular reduction. There are no load-bearing self-citations and no imported uniqueness theorem.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central method rests on the assumption that a single linear shift, fit to the difference between generated captions and original descriptions, captures abstractness rather than captioning style, and that this shift transfers across datasets and LLM-rewritten queries. None of the entries are invented physical entities; the free parameters are the PCA rank k, the concreteness threshold, and the query standardization statistics, all of which influence the reported numbers.

free parameters (3)
  • k (number of PCA components) = 600
    Sec. 4 states k=600 is an empirical choice; no sensitivity analysis is reported, and retrieval gains depend on this value.
  • concreteness threshold = 3.0
    Sec. 3 classifies adjectives as abstract if their Brysbaert concreteness rating is below 3.0; this threshold is from [5] and determines all abstract/concrete statistics.
  • query standardization statistics mu_q, sigma_q = not reported
    Eq. (8) standardizes query representations using 'the mean and standard deviation among query representations'; the source (training, test, or batch) is unspecified, and if test queries are used this is a transductive design choice.
assumptions (5)
  • domain assumption The Brysbaert concreteness lexicon with threshold 3.0 reliably distinguishes abstract from concrete fashion adjectives after manual overrides for context-dependent words such as 'cool' and 'sharp'.
    Supp. B.2 details manual overrides; the classification underpins all statistics in Sec. 3.
  • ad hoc to paper The embedding difference h_C - h_A between captioning-model captions and original fashion descriptions is dominated by the abstract-to-concrete distinction and is well captured by a linear PCA shift that transfers to new queries.
    This is the core assumption behind Eqs. (5)-(8); if the difference reflects captioning style, length, or vocabulary instead of abstractness, ACT is a caption-style shift, not an abstract-to-concrete translator.
  • ad hoc to paper LLM rewriting makes abstract queries more concrete without discarding retrieval-relevant information.
    Inference in Sec. 4 assumes q' can be shifted with the same A-C statistics; the FACAD cross-dataset results in Supp. C.2 show rewriting can hurt, so this assumption is not universally valid.
  • domain assumption spaCy POS tagging and the hand-written attribute extraction rules correctly identify adjective-noun attributes in fashion text.
    All dataset statistics in Sec. 3 and Supp. B.2 rely on this pipeline.
  • domain assumption The captioning models (Qwen2-VL, CogVLM2) produce visually grounded, concrete-oriented captions.
    The A-C database construction in Sec. 4 assumes generated captions are concrete counterparts; Tab. 2 supports this for the tested models but it depends on the captioner.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Seeing the Abstract: Translating the Abstract Language for Vision Language Models." pith.science (2026). https://pith.science/paper/A4ETHOAO

@misc{pith2026250503242,
  author       = {Pith},
  title        = {Pith review of: Seeing the Abstract: Translating the Abstract Language for Vision Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/A4ETHOAO}},
  note         = {Machine review of arXiv:2505.03242}
}
read the original abstract

Natural language goes beyond dryly describing visual content. It contains rich abstract concepts to express feeling, creativity and properties that cannot be directly perceived. Yet, current research in Vision Language Models (VLMs) has not shed light on abstract-oriented language. Our research breaks new ground by uncovering its wide presence and under-estimated value, with extensive analysis. Particularly, we focus our investigation on the fashion domain, a highly-representative field with abstract expressions. By analyzing recent large-scale multimodal fashion datasets, we find that abstract terms have a dominant presence, rivaling the concrete ones, providing novel information, and being useful in the retrieval task. However, a critical challenge emerges: current general-purpose or fashion-specific VLMs are pre-trained with databases that lack sufficient abstract words in their text corpora, thus hindering their ability to effectively represent abstract-oriented language. We propose a training-free and model-agnostic method, Abstract-to-Concrete Translator (ACT), to shift abstract representations towards well-represented concrete ones in the VLM latent space, using pre-trained models and existing multimodal databases. On the text-to-image retrieval task, despite being training-free, ACT outperforms the fine-tuned VLMs in both same- and cross-dataset settings, exhibiting its effectiveness with a strong generalization capability. Moreover, the improvement introduced by ACT is consistent with various VLMs, making it a plug-and-play solution.

Figures

Figures reproduced from arXiv: 2505.03242 by the authors.

Figure 1
Figure 1. Human language can exhibit both abstract and concrete words to express feelings, desires, and properties to￾gether with perceivable elements, e.g., when describing a fash￾ion item. However, Vision Language Models (VLMs) are mostly pre-trained with concrete -oriented web-image texts, thus under￾representing the abstract-oriented ones. When encoding the abstract-oriented description with pre-trained VLMs, there exists… view at source ↗
Figure 2
Figure 2. Top: Wordcloud of the abstract and concrete ad￾jectives in the DeepFashion dataset. The larger font represents a higher frequency. Bottom: Distribution of maximum absolute Matthews’s Correlation Coefficient (MCC) between each abstract attribute and concrete ones in DeepFashion. The peak near 0 re￾veals that the majority of abstract attributes have a low correlation with the concrete ones. tion (median 37 words) orig… view at source ↗
Figure 3
Figure 3. Left: Retrieval performance of an ideal system on Deepfashion original descriptions when the majority of present at￾tributes are concrete , abstract or mixed . Abstract attributes allow for better retrieval performance. Right: performance of current VLMs on DeepFashion when using original abstract de￾scriptions or concrete VLM generated ones. Current VLMs achieve better performance with concrete-oriented description… view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Overview of our two-phase Abstract-to-Concrete Translator (ACT). During the [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 1
Figure 1. Figure 1: Samples taken from the datasets considered for analy [PITH_FULL_IMAGE:figures/full_fig_p011_1.png]
Figure 2
Figure 2. Figure 2: Wordcloud of concrete (blue) and abstract (red) ad￾jectives in FACAD descriptions. The larger font indicates a higher frequency. Abstract adjectives make up a large portion of the en￾tire dataset, being as frequent as concrete ones (see Tab. 1 of the main paper). extra…
Figure 3
Figure 3. Figure 3: Wordcloud of concrete (blue) and abstract (red) ad￾jectives in LAION descriptions. The larger font indicates a higher frequency. It’s clear how concrete adjectives are more frequent than abstract ones (see Tab. 1 of the main paper). choice is guided by our observation …
Figure 4
Figure 4. Figure 4: Density distribution of the cosine similarity with respect [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: t-SNE of VLM latent space for textual embeddings in the evaluation set of DeepFashion. Connected markers denote different [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]
Figure 6
Figure 6. Figure 6: Some failure cases of ACT on FACAD. On the left, we report failures due to the generic FACAD descriptions: despite ACT [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: Qualitative examples of retrieval using our ACT, on the test split of DeepFashion, using abstract attributes. In this example, we [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]
Figure 8
Figure 8. Figure 8: Further qualitative examples of retrieval using our ACT, on the test split of DeepFashion, using abstract attributes. In the top [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]
Figure 9
Figure 9. Figure 9: Further retrieval results from our ACT on the test split of DeepFashion showcase that ACT can also interpret abstract global [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 32 canonical work pages

  1. [1]

    Phi-3 technical report: A highly capable language model locally on your phone

    Marah Abdin, Jyoti Aneja, Hany Awadalla, Ahmed Awadal- lah, Ammar Ahmad Awan, Nguyen Bach, Amit Bahree, Arash Bakhtiari, Jianmin Bao, Harkirat Behl, et al. Phi-3 technical report: A highly capable language model locally on your phone. arXiv preprint arXiv:2404.14219, 2024. 3

  2. [2]

    Fashion product images (small), 2019

    Param Aggarwal. Fashion product images (small), 2019. Ac- cessed: March 1, 2024. 2

  3. [3]

    Compositional learning of image-text query for image retrieval

    Muhammad Umer Anwaar, Egor Labintcev, and Martin Kle- insteuber. Compositional learning of image-text query for image retrieval. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2021. 3

  4. [4]

    Effective conditioned and composed im- age retrieval combining clip-based features

    Alberto Baldrati, Marco Bertini, Tiberio Uricchio, and Al- berto Del Bimbo. Effective conditioned and composed im- age retrieval combining clip-based features. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, 2022. 3

  5. [5]

    Concreteness ratings for 40 thousand generally known en- glish word lemmas

    Marc Brysbaert, Amy Beth Warriner, and Victor Kuperman. Concreteness ratings for 40 thousand generally known en- glish word lemmas. Behavior research methods, 2014. 1, 3, 2

  6. [6]

    Openfash- ionclip: Vision-and-language contrastive learning with open- source fashion data

    Giuseppe Cartella, Alberto Baldrati, Davide Morelli, Mar- cella Cornia, Marco Bertini, and Rita Cucchiara. Openfash- ionclip: Vision-and-language contrastive learning with open- source fashion data. In Proceedings of International Confer- ence on Image Analysis and Processing, 2023. 2, 3, 5, 7

  7. [7]

    Reproducible scal- ing laws for contrastive language-image learning

    Mehdi Cherti, Romain Beaumont, Ross Wightman, Mitchell Wortsman, Gabriel Ilharco, Cade Gordon, Christoph Schuh- mann, Ludwig Schmidt, and Jenia Jitsev. Reproducible scal- ing laws for contrastive language-image learning. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023. 3, 7

  8. [8]

    Contrastive language and vi- sion learning of general fashion concepts

    Patrick John Chia, Giuseppe Attanasio, Federico Bianchi, Silvia Terragni, Ana Rita Magalh˜aes, Diogo Goncalves, Ciro Greco, and Jacopo Tagliabue. Contrastive language and vi- sion learning of general fashion concepts. Scientific Reports,

Show all 43 references
  1. [9]

    Style finder: Fine-grained clothing style detection and retrieval

    Wei Di, Catherine Wah, Anurag Bhardwaj, Robinson Pira- muthu, and Neel Sundaresan. Style finder: Fine-grained clothing style detection and retrieval. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2013. 3

  2. [10]

    A survey on in-context learning

    Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia, Jingjing Xu, Zhiyong Wu, Tianyu Liu, et al. A survey on in-context learning. arXiv preprint arXiv:2301.00234, 2022. 3

  3. [11]

    The llama 3 herd of models

    Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Ab- hishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783 ,

  4. [12]

    Dat- acomp: In search of the next generation of multimodal datasets

    Samir Yitzhak Gadre, Gabriel Ilharco, Alex Fang, Jonathan Hayase, Georgios Smyrnis, Thao Nguyen, Ryan Marten, Mitchell Wortsman, Dhruba Ghosh, Jieyu Zhang, et al. Dat- acomp: In search of the next generation of multimodal datasets. Advances in Neural Information Processing Sys...

  5. [13]

    Fashionvlp: Vision language transformer for fashion re- trieval with feedback

    Sonam Goenka, Zhaoheng Zheng, Ayush Jaiswal, Rakesh Chada, Yue Wu, Varsha Hedau, and Pradeep Natarajan. Fashionvlp: Vision language transformer for fashion re- trieval with feedback. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, 2022. 3

  6. [14]

    Scott, and Serge Belongie

    Sheng Guo, Weilin Huang, Xiao Zhang, Prasanna Srikhanta, Yin Cui, Yuan Li, Hartwig Adam, Matthew R. Scott, and Serge Belongie. The imaterialist fashion attribute dataset. In Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops, 2019. 2, 3

  7. [15]

    Huang, Xiao Zhang, Menglong Zhu, Yuan Li, Yang Zhao, and Larry S

    Xintong Han, Zuxuan Wu, Phoenix X. Huang, Xiao Zhang, Menglong Zhu, Yuan Li, Yang Zhao, and Larry S. Davis. Automatic spatially-aware fashion concept discovery. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 2017. 2, 3

  8. [16]

    Cogvlm2: Visual language mod- els for image and video understanding

    Wenyi Hong, Weihan Wang, Ming Ding, Wenmeng Yu, Qingsong Lv, Yan Wang, Yean Cheng, Shiyu Huang, Jun- hui Ji, Zhao Xue, et al. Cogvlm2: Visual language mod- els for image and video understanding. arXiv preprint arXiv:2408.16500, 2024. 8, 2

  9. [17]

    spacy: Industrial-strength natural lan- guage processing in python, 2020

    Matthew Honnibal, Ines Montani, Sofie Van Landeghem, and Adriane Boyd. spacy: Industrial-strength natural lan- guage processing in python, 2020. 3, 2

  10. [18]

    Feris, Qiang Chen, and Shuicheng Yan

    Junshi Huang, Rogerio S. Feris, Qiang Chen, and Shuicheng Yan. Cross-domain image retrieval with a dual attribute- aware ranking network. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 2015. 3

  11. [19]

    Multi-label fashion image classification with minimal human supervision

    Naoto Inoue, Edgar Simo-Serra, Toshihiko Yamasaki, and Hiroshi Ishikawa. Multi-label fashion image classification with minimal human supervision. In Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops, 2017. 2

  12. [20]

    Cross- domain image retrieval with attention modeling

    Xin Ji, Wei Wang, Meihui Zhang, and Yang Yang. Cross- domain image retrieval with attention modeling. InProceed- ings of ACM International Conference on Multimedia, 2017. 3

  13. [21]

    Hadi Kiapour, Xufeng Han, Svetlana Lazebnik, Alexan- der C

    M. Hadi Kiapour, Xufeng Han, Svetlana Lazebnik, Alexan- der C. Berg, and Tamara L. Berg. Where to buy it: Match- ing street clothing photos in online shops. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, 2015. 3

  14. [22]

    Cosmo: Content-style modulation for image retrieval with text feed- back

    Seungmin Lee, Dongwan Kim, and Bohyung Han. Cosmo: Content-style modulation for image retrieval with text feed- back. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 2021. 3

  15. [23]

    Mind the gap: Understanding the modality gap in multi-modal contrastive representation learning

    Victor Weixin Liang, Yuhui Zhang, Yongchan Kwon, Ser- ena Yeung, and James Y Zou. Mind the gap: Understanding the modality gap in multi-modal contrastive representation learning. Advances in Neural Information Processing Sys- tems, 2022. 8

  16. [24]

    Deepfashion: Powering robust clothes recog- nition and retrieval with rich annotations

    Ziwei Liu, Ping Luo, Shi Qiu, Xiaogang Wang, and Xi- aoou Tang. Deepfashion: Powering robust clothes recog- nition and retrieval with rich annotations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2016. 2, 3, 7

  17. [25]

    Matthews

    Brian W. Matthews. Comparison of the predicted and ob- served secondary structure of t4 phage lysozyme. Biochim- ica et Biophysica Acta (BBA)-Protein Structure, 1975. 4

  18. [26]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision. In Proceedings of In...

  19. [27]

    Fashion-gen: The generative fashion dataset and challenge

    Negar Rostamzadeh, Seyedarian Hosseini, Thomas Boquet, Wojciech Stokowiec, Ying Zhang, Christian Jauvin, and Chris Pal. Fashion-gen: The generative fashion dataset and challenge. arXiv preprint arXiv:1806.08317, 2018. 2, 3

  20. [28]

    Laion-400m: Open dataset of clip-filtered 400 million image-text pairs

    Christoph Schuhmann, Richard Vencu, Romain Beaumont, Robert Kaczmarczyk, Clayton Mullis, Aarush Katta, Theo Coombes, Jenia Jitsev, and Aran Komatsuzaki. Laion-400m: Open dataset of clip-filtered 400 million image-text pairs. arXiv preprint arXiv:2111.02114, 2021. 2, 3

  21. [29]

    A tutorial on principal component analysis

    Jonathon Shlens. A tutorial on principal component analysis. arXiv preprint arXiv:1404.1100, 2014. 5

  22. [30]

    Neuroaesthetics in fashion: Mod- eling the perception of fashionability

    Edgar Simo-Serra, Sanja Fidler, Francesc Moreno-Noguer, and Raquel Urtasun. Neuroaesthetics in fashion: Mod- eling the perception of fashionability. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2015. 2

  23. [31]

    Eva-clip: Improved training techniques for clip at scale

    Quan Sun, Yuxin Fang, Ledell Wu, Xinlong Wang, and Yue Cao. Eva-clip: Improved training techniques for clip at scale. arXiv preprint arXiv:2303.15389, 2023. 3, 7

  24. [32]

    What makes a style: Experimental analysis of fashion prediction

    Moeko Takagi, Edgar Simo-Serra, Satoshi Iizuka, and Hi- roshi Ishikawa. What makes a style: Experimental analysis of fashion prediction. InProceedings of the IEEE/CVF Inter- national Conference on Computer Vision Workshops, 2017. 2

  25. [33]

    Visualizing data using t-sne

    Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of machine learning research , 9 (11), 2008. 5

  26. [34]

    Vinson, Marco Tettamanti, Joseph T

    Gabriella Vigliocco, Stavroula-Thaleia Kousta, Pasquale An- thony Della Rosa, David P. Vinson, Marco Tettamanti, Joseph T. Devlin, and Stefano F. Cappa. The neural rep- resentation of abstract words: the role of emotion. Cerebral Cortex, 2014. 1

  27. [35]

    Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution

    Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, et al. Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution. arXiv preprint arXiv:2409.12191, 2024. 7, 2

  28. [36]

    Clothes search in con- sumer photos via color matching and attribute learning

    Xianwang Wang and Tong Zhang. Clothes search in con- sumer photos via color matching and attribute learning. In Proceedings of ACM International Conference on Multime- dia, 2011. 3

  29. [37]

    Fashion iq: A new dataset towards retrieving images by natural language feedback

    Hui Wu, Yupeng Gao, Xiaoxiao Guo, Ziad Al-Halah, Steven Rennie, Kristen Grauman, and Rogerio Feris. Fashion iq: A new dataset towards retrieving images by natural language feedback. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021. 2, 3

  30. [38]

    Fashion- mnist: a novel image dataset for benchmarking machine learning algorithms

    Han Xiao, Kashif Rasul, and Roland V ollgraf. Fashion- mnist: a novel image dataset for benchmarking machine learning algorithms. arXiv preprint arXiv:1708.07747, 2017. 2

  31. [39]

    Fashion captioning: Towards generating accurate descrip- tions with semantic rewards

    Xuewen Yang, Heming Zhang, Di Jin, Yingru Liu, Chi-Hao Wu, Jianchao Tan, Dongliang Xie, Jue Wang, and Xin Wang. Fashion captioning: Towards generating accurate descrip- tions with semantic rewards. In Proceedings of European Conference on Computer Vision, 2020. 2, 3, 7

  32. [40]

    Sigmoid loss for language image pre-training

    Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. Sigmoid loss for language image pre-training. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023. 7

  33. [41]

    chic” and “street ready

    Mingchen Zhuge, Dehong Gao, Deng-Ping Fan, Linbo Jin, Ben Chen, Haoming Zhou, Minghui Qiu, and Ling Shao. Kaleido-bert: Vision-language pre-training on fashion do- main. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 2021. 3 Seeing the ...

  34. [42]

    adjective

    If the head is either an “adjective”, or was tagged as the “subject” of the sentence, lemmatize it and create the attribute (token, head)

  35. [43]

    compound word

    If the token was a “compound word” ( i.e., a concept ex- pressed through multiple words, such as “knee length”), extract all the (token, compound) couples as attributes. iv) Classification phase. Finally, we classify the attributes following the methodology and lexicon present...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.