Improved Feature Generating Framework for Transductive Zero-shot Learning

Kaizhu Huang; Shiming Chen; Xiaobo Jin; Xinyuan Ru; Yaochu Jin; Zihan Ye

REVIEW 4 major objections 5 minor 45 references

Improved Feature Generating Framework for Transductive Zero-shot Learning

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read The paper proves that a tiny unseen-class prior bias can sharply cut transductive zero-shot accuracy, and replaces the flawed discriminator with pseudo-conditional training.

desk verdict Useful empirical diagnosis of where prior bias hits TZSL, but the proposed fix's central claim—that PFA circumvents prior estimation—doesn't survive contact with its own training procedure. read the letter →

arxiv 2412.18282 v1 pith:5M5XWMAW submitted 2024-12-24 cs.CV

Zihan Ye , Xinyuan Ru , Shiming Chen , Yaochu Jin , Kaizhu Huang , Xiaobo Jin This is my paper · ORCID

classification cs.CV

keywords transductivezero-shotlearningfeaturegenerationunseen-classpriorbiaspseudo-conditionaladversarialvariationalembeddingregressionsemanticGAN

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

Transductive zero-shot learning lets a model train on unlabeled examples of classes it has never seen, so the assumed prior over those unseen classes matters. This paper tries to establish that even a small error in that prior causes a large accuracy drop, and that the drop is caused by the unconditional unseen discriminator used in standard feature-generating frameworks, not by the generator or classifier alone. The paper proves that at the global optimum of the adversarial game, when the unconditional generated distribution matches the real one, the per-class distributions still differ by an accumulated prior error. It then proposes a framework that conditions the unseen discriminator on pseudo-labels from a semantic regressor, bypassing the need to estimate the prior, and augments the regressor with a variational embedding pretraining step. A sympathetic reader would care because this explains a known fragility of transductive zero-shot models and offers a concrete architectural substitute.

What carries the argument

The load-bearing component is the unconditional unseen discriminator $D_u$ from the f-VAEGAN baseline, which judges real versus synthesized unseen features without a class condition; the paper proves that this design makes the per-class gap unavoidable under a wrong prior. Its Accumulated Prior Error (APE) proposition states that the per-class gap can be written as $e(x^u, y_i^u) = \frac{|p_r(y_i^u|x^u)p_g(y_i^u)-p_g(y_i^u|x^u)p_r(y_i^u)|}{p_r(y_i^u)p_g(y_i^u)}p_r(x^u)$, which vanishes only when the posterior ratios align with the prior ratios. PFA replaces $D_u$ with a conditional discriminator $D_{u2}$ that takes the predicted pseudo condition $\tilde{a}^u=R(x^u)$ as an extra input, and VER trains an unsupervised VAE to produce embeddings $[\mu^{pre},\sigma^{pre}]$ that are concatenated with visual features to improve $R$. The mechanism works by matching real and generated samples by class rather than by unconditional statistics, which removes the need for prior estimation.

What would settle it

Train the proposed framework on AWA1 or AWA2 while artificially corrupting the pseudo-conditions on unseen classes, for example by replacing $\tilde{a}^u=R(x^u)$ with noise or with the output of a deliberately biased regressor; if accuracy does not drop as regressor error rises, the claim that PFA works through accurate pseudo-conditions is wrong. A supporting observation would be a monotone accuracy-versus-regressor-error curve, and the paper does not report that curve.

Watch

Extended reading notes

Core claim

The central discovery is a 'prior reaction chain': the assumed prior on unseen classes enters through the unconditional discriminator, which accumulates the bias and misleads the generator; at the global optimum of the minimax game, with $p_r(x^u)=p_g(x^u)$, the class-conditioned generation distributions still differ, with the gap quantified by the Accumulated Prior Error identity of Proposition 3.1. The paper shows that conditioning the unseen discriminator on pseudo class-conditions predicted by a regressor avoids this chain, because the pseudo-conditions carry the real class information once the regressor is accurate. The resulting I-VAEGAN combines Pseudo-conditional Feature Adversarial learning with Variational Embedding Regression and reports the best transductive zero-shot and generalized transductive zero-shot accuracies on AWA1, AWA2, CUB, and SUN across the tested priors.

Load-bearing premise

The whole remedy rests on the semantic regressor being accurate enough on unseen classes; if the pseudo-conditions are poor, the conditional discriminator can reinforce the wrong class assignments just as a biased prior would.

Editorial extensions

If this is right

Any f-VAEGAN-style transductive zero-shot method that keeps an unconditional unseen discriminator remains vulnerable to prior bias, even with a perfect generator at the unconditional level.
Using pseudo-conditions predicted from real unseen samples lets the model skip prior estimation; under uniform, CPE, and ground-truth priors, the proposed framework outperforms its predecessor on all four reported benchmarks.
The variational embedding regression is a drop-in addition: attaching it to other regressors (TF-VAEGAN and FREE) lowers semantic regression error on unseen classes.
The accuracy gain is largest on datasets with non-uniform unseen class distributions, which is where the prior reaction chain does the most damage.
With an accurate enough regressor, pseudo-conditions approximate ground-truth class conditions, collapsing the prior reaction chain and reducing the Accumulated Prior Error.

Reading between the lines

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

An unstated diagnostic follows from the paper's APE identity: per-class or per-dataset APE could serve as a monitoring metric for transductive zero-shot models, independent of the specific architecture; the paper does not propose it as a general evaluation tool.
Because the guarantee is conditional on regressor accuracy, a natural extension is to couple PFA with uncertainty estimates on the pseudo-conditions and down-weight low-confidence ones; the paper does not explore that.
The same 'unconditional discriminator accumulates class-prior bias' mechanism may apply to other generative settings with imbalanced unlabeled data, such as semi-supervised generation or unsupervised domain adaptation; the paper only tests it on zero-shot benchmarks.
Deriving a quantitative bound from regressor error to final classification error would turn the sufficient-condition argument into a provable robustness guarantee; this paper leaves that bound open.

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.

Desk Editor's Note

read the letter

Here's my read. The genuinely new thing is Table 2: the controlled experiment separating the prior used to train Du from the prior used to train G. That's a clean, informative diagnostic, and it makes a real point—the unconditional discriminator is the sensitive link in the prior reaction chain. The APE proposition itself is just Bayes' rule, but framing it as an accumulated prior error is fine; the math is correct, if not deep. VER is a simple unsupervised VAE pretraining that seems to give consistent, if modest, regression improvements and transfers to other methods (Fig. 8). The accuracy gains over Bi-VAEGAN are small, 0.1–2.4 points, and the tables honestly show that. Now the soft spots, proportionate to their size. The stress-test concern lands. PFA's whole selling point is that it "circumvents the need for prior estimation" by conditioning on pseudo-semantics from R. But R is trained in Stage 2 with adversarial labels sampled according to the very prior being circumvented (Algorithm 1, line 14; Eq. 26). If the prior is biased, the bias doesn't disappear; it moves into R, and from there into the pseudo-conditions feeding Du2 and G. The paper gives no bound relating R's unseen-class regression error to APE or T1, and the AWA1 uniform-prior result (67.0 vs. 94.4 with GT) is consistent with R failing to correct the bias. The claim that "once our R is accurate enough" the pseudo-conditions approximate ground truth is an assumption, not a mechanism, and the training objective doesn't ensure it. That's a load-bearing overstatement, though not a fatal one—the method still works reasonably, just not for the stated reason. Also worth noting: Proposition 3.1's "inevitable" depends on the GAN reaching a global optimum, which never happens in practice. That's a standard crutch, but the wording overreaches. Reproducibility is a moderate issue: no code, no error bars, and the hyperparameter reporting is incomplete (λu2 is given, but λu1 and λr are not; exact values for nsyn per dataset appear only in figures). Who is this for? People working on generative TZSL will get a useful diagnostic and a decent baseline extension. I'd probably cite the prior-sensitivity experiment if I worked in the area. It deserves a serious referee—the question of where prior bias enters the pipeline is worth asking—but the reviewers should push hard on the pseudo-condition mechanism and request code, error bars, and a toned-down abstract. I'd send it out, but with the expectation of heavy revision.

Referee Report

4 major / 5 minor

Summary. The paper studies transductive zero-shot learning (TZSL) with generative feature models, focusing on the effect of the unseen-class prior. It reports that small prior bias causes large accuracy drops (Table 1), that the unconditional unseen discriminator D_u is the sensitive component (Table 2), and that even at the global optimum p_r(x_u)=p_g(x_u), class-conditional generation distributions remain mismatched when the estimated prior differs from the true prior (Proposition 3.1, Appendix A). To address this, the paper proposes I-VAEGAN with two components: Pseudo-conditional Feature Adversarial learning (PFA), which uses a semantic regressor R to produce pseudo class conditions ã_u = R(x_u) and feeds them to a conditional discriminator D_u2, and Variational Embedding Regression (VER), which pre-trains a VAE on visual features and concatenates the variational embedding to the regressor input. Experiments on AWA1, AWA2, CUB, and SUN compare TZSL and TGZSL accuracy under uniform, CPE, and ground-truth priors, showing consistent improvements over the Bi-VAEGAN baseline and state-of-the-art results in several settings.

Significance. If the prior-reaction-chain diagnosis is correct, the paper identifies a previously under-appreciated cause of TZSL degradation under imperfect unseen-class priors and provides a practical fix that does not require accurate prior estimation. The controlled experiments in Tables 1 and 2 are well designed and support the sensitivity of D_u to prior bias. The proposed PFA and VER are simple, plug-in components that could transfer to other generative TZSL frameworks; the VER transfer experiments on TF-VAEGAN and FREE in Appendix C.5 support this. The theoretical Proposition 3.1 is algebraically correct and makes explicit the class-conditional gap that remains at the unconditional GAN optimum. The paper provides a clear overall architecture and detailed training algorithm (Algorithm 1). However, the strength of the central claim—that PFA circumvents the need for prior estimation—depends on the accuracy of the semantic regressor R on unseen classes, and the paper does not provide a formal or empirical bound on that accuracy or on how regressor error translates to APE or classification accuracy.

major comments (4)

[Sec. 3.3 and Algorithm 1, lines 14-18] The claim that PFA 'circumvents the need for prior estimation' is only partially supported. The regressor R is trained in Stage-2 using real unseen-class semantics a_u sampled according to the very prior that PFA is meant to avoid (Algorithm 1 line 14; Eq. 26). If that prior is biased, the adversarial regressor Dr pushes R to produce pseudo-semantics whose class distribution matches the wrong prior, so the prior bias is relocated into R and then into the pseudo-conditions ã_u. The paper acknowledges this only through the statement that 'once our R is accurate enough' the pseudo conditions approximate ground truth, but no bound is given relating R's unseen-class regression error to APE or to final accuracy. The empirical results with uniform prior on AWA1 (Table 3: T1=67.0 vs 94.4 with GT prior) are consistent with R failing to supply correct class conditions under a biased prior. The central claim of circumventing prior estimation should be either substantiated with an error-bound analysis or softened to 'reduces sensitivity to prior bias'.
[Proposition 3.1 and Remark 3.2] The statement that the class-specific gap is 'inevitable' at the global optimum is correct only under the assumed prior p_g(y_u) that is used to sample unseen classes during GAN training. The proposition is a definition of APE followed by Bayes' rule, and the remark correctly identifies the condition under which the gap vanishes. However, the paper's empirical claim that this gap is large in practice (Fig. 4) is not connected to the proposition in a quantitative way: the APE shown is computed from the trained classifier, but no analysis is provided of whether the observed APE is driven by the prior bias or by other factors such as finite-sample estimation or the classifier's own error. As a result, the causal chain pg(y_u) → D_u → G → f_zsl is not established with the same rigor as the algebraic identity; it is supported only by the controlled experiments in Table 2, which are suggestive but not definitive.
[Eq. (8) and Eq. (30)] The PFA objective uses the same pseudo-condition ã_u for both the real sample x_u and the generated sample ã_x_u. If R is imperfect, this can cause the discriminator to associate the wrong class condition with real samples, potentially reinforcing incorrect class assignments during generation. The paper does not analyze how regressor error interacts with the gradient penalty term or with the generator update. A concrete test would be to measure the classification error of x_u under the pseudo-conditions versus the true conditions, on datasets where the true condition is available for validation, and to report the relation between that error and the final T1 accuracy.
[Table 6 and Table 8] The ablation study in Table 6 shows that PFA alone (with CPE prior) improves AWA1 from 90.0 to 91.3, but the improvement is much smaller than the gap between CPE and GT prior (93.9 for the baseline with GT). Table 8 reports PFA under different priors but uses a different evaluation (visual-space GZSL) and shows mixed results: on AWA1, PFA with CPE prior improves H from 44.3 to 48.4, but on AWA2 with GT prior PFA slightly decreases H (53.7 to 53.1) and on CUB with uniform prior PFA improves H while decreasing U. These inconsistencies are not discussed; the claim that PFA uniformly mitigates prior bias would be strengthened by a more complete analysis of when PFA helps and when it hurts.

minor comments (5)

[Abstract and Sec. 1] The abstract states that I-VAEGAN 'achieves state-of-the-art TZSL accuracy across various benchmarks and priors', but the gains over Bi-VAEGAN in Table 4 are modest (e.g., AWA2 T1 95.9 vs 95.8, CUB T1 77.2 vs 76.8). The phrase 'state-of-the-art' should be qualified, or the comparisons should include confidence intervals to demonstrate that the improvements are statistically significant.
[Notation, Sec. 3.1.1] The notation for unseen semantics uses A_u both as the set of class-level semantic labels and as the set of instance-level labels; the sentence 'we can use unseen semantic labels As = {au_i}_{i=1}^{Nu}' contains a typo (As should be Au). Clarify the distinction between class-level and instance-level semantics.
[Algorithm 1, line 26] In Algorithm 1, line 26, the latent variable is sampled as 'tilde z_s ∼ N(µ^pre, σ^pre)' but should be 'N(µ^s, σ^s)' according to Eq. (1). This appears to be a typographical error and should be corrected.
[Sec. 3.4 and Eq. (11)] The VAE reconstruction loss in Eq. (11) uses a minus sign before the expectation, which is unconventional; the standard VAE loss uses a plus sign (the negative log-likelihood term). The sign convention should be explained or corrected.
[Table 3 and Sec. 4.2.2] The comparison under CPE prior is only against Bi-VAEGAN, while the uniform prior comparison includes f-VAEGAN and TF-VAEGAN. For a fair evaluation of the claim that PFA improves over other methods under imperfect priors, the authors should also evaluate f-VAEGAN and TF-VAEGAN with CPE prior, or explain why this is not feasible.

Circularity Check

1 steps flagged · score 4.0 of 10

PFA's 'circumvent prior estimation' claim is undercut because the regressor supplying pseudo-conditions is itself trained on prior-sampled unseen semantics; benchmark accuracy comparisons remain independent.

fitted input called prediction [Sec. 3.3 (PFA), Eqs. (8), (12), (26), (30); Appendix B.2, Algorithm 1 Stage-2 line 14]
"PFA circumvents the need for prior estimation by explicitly injecting the predicted semantics as pseudo conditions for unseen classes premised by precise semantic regression. ... Algorithm 1, Stage-2: 'Sample a batch of semantic labels au according to predefined prior estimation;' ... Lu2_GAN = E[Du2(xu, ˜au)] − E[Du2(˜xu, ˜au)] (Eq. 30)."

The paper claims PFA removes the need for prior estimation because pseudo-conditions ãu = R(xu) match the ground-truth prior if R is accurate. But by the paper's own training procedure, R is fit in Stage-2 with the adversarial loss Lr_GAN (Eq. 26), where the real unseen semantics au are sampled from the predefined prior estimation (Algorithm 1, line 14). Dr therefore drives R to produce ãu whose class distribution matches that estimated prior, not the per-instance ground-truth classes. Stage-3's Du2 then conditions on these ãu (Eq. 30). Consequently, any prior bias is not circumvented; it is transferred into R and thence into the pseudo-conditions. The paper supplies no bound under which R's unseen-class regression error becomes small, and its uniform-prior AWA1 result (T1=67.0 vs.

full rationale

Proposition 3.1/APE is a Bayes-rule identity: under pr(xu)=pg(xu), the stated expression for e is exactly |pr(xu|yi)-pg(xu|yi)|. That is a definitional rewriting rather than a circular derivation of the paper's diagnosis; the diagnosis also rests on controlled experiments (Table 2) and on the gap remaining nonzero in practice. The method's main empirical support is standard T1/H accuracy on AWA1/AWA2/CUB/SUN against published baselines (Tables 3-5, 7-8), which is external to the paper's fitted quantities. The one substantial circularity-adjacent flaw is PFA's claim to circumvent prior estimation: the pseudo-conditions are produced by R, but R is adversarially trained in Stage-2 against unseen semantics au sampled from the very prior being circumvented (Algorithm 1 line 14, Eq. 26). Hence the prior enters through R and then through Du2's conditioning (Eq. 30); the claim that accurate R makes pseudo-conditions prior-free is an unproven premise, not a consequence of the objective. In addition, the APE reduction shown in Fig. 4 is partly self-referential because APE is the quantity PFA was designed to minimize, though the paper does not rely on it as the sole evidence. Overall, the benchmark results give the central claim independent content, so a moderate score is appropriate.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central new machinery (PFA) requires the existence of an accurate semantic regressor for unseen classes; the paper does not prove or quantify this, it is a domain assumption. The adversarial/VAE training relies on standard GAN and VAE theory. The main free parameters are loss weights and synthesis counts, tuned per dataset.

free parameters (4)

lambda_u2 = 0.09
Loss weight for the pseudo-conditional adversarial loss Lu2_GAN; tuned on validation and reported as optimal in Appendix C.3.2.
n_syn = 3000 for AWA1/AWA2, 120 for CUB, 400 for SUN
Number of synthetic unseen-class features generated at inference; chosen by hand per dataset (Appendix C.3.2).
lambda_u1 = not reported
Weight for the original unconditional adversarial loss Lu1_GAN; mentioned in Eq. 12 but its value is not given in the paper.
lambda_r = not reported
Weight for adversarial regression loss Lr_GAN; specified as a hyperparameter but no value stated.

assumptions (4)

standard math GAN global optimum implies pr(x)=pg(x)
Invoked in the proof of Proposition 3.1. This is standard GAN theory from Goodfellow et al., 2014.
domain assumption Semantic regressor can be accurate on unseen classes
The PFA method requires that R predicts semantics well enough for unseen classes; stated in Sec 3.3 without proof or error bounds.
domain assumption Pre-extracted visual features contain enough semantic information for regression
Standard in ZSL; features from ResNet101 pretrained on ImageNet are used directly.
domain assumption Unseen test samples are available unlabeled during training
This is the defining assumption of transductive ZSL, used throughout the method.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Improved Feature Generating Framework for Transductive Zero-shot Learning." pith.science (2026). https://pith.science/paper/5M5XWMAW

@misc{pith2026241218282,
  author       = {Pith},
  title        = {Pith review of: Improved Feature Generating Framework for Transductive Zero-shot Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5M5XWMAW}},
  note         = {Machine review of arXiv:2412.18282}
}

read the original abstract

Feature Generative Adversarial Networks have emerged as powerful generative models in producing high-quality representations of unseen classes within the scope of Zero-shot Learning (ZSL). This paper delves into the pivotal influence of unseen class priors within the framework of transductive ZSL (TZSL) and illuminates the finding that even a marginal prior bias can result in substantial accuracy declines. Our extensive analysis uncovers that this inefficacy fundamentally stems from the utilization of an unconditional unseen discriminator - a core component in existing TZSL. We further establish that the detrimental effects of this component are inevitable unless the generator perfectly fits class-specific distributions. Building on these insights, we introduce our Improved Feature Generation Framework, termed I-VAEGAN, which incorporates two novel components: Pseudo-conditional Feature Adversarial (PFA) learning and Variational Embedding Regression (VER). PFA circumvents the need for prior estimation by explicitly injecting the predicted semantics as pseudo conditions for unseen classes premised by precise semantic regression. Meanwhile, VER utilizes reconstructive pre-training to learn class statistics, obtaining better semantic regression. Our I-VAEGAN achieves state-of-the-art TZSL accuracy across various benchmarks and priors. Our code would be released upon acceptance.

Figures

Figures reproduced from arXiv: 2412.18282 by the authors.

**Figure 1.** Our discovered prior reaction chain. We find that unseen prior probabilities firstly impact unconditional unseen discriminator Du. Then, Du gives insufficient gradient guidance to generator G. Next, although unconditional generation distribution pg(x u ) could fully fit to real unconditional pr(x u ), class-specific distributions still have an inevitable gap. Finally, the ZSL classifier mis-classifies test samples. … view at source ↗

**Figure 2.** Illustration of our proposed PFA. (a) Standard unconditional feature adversarial learning can be negatively affected by prior bias and unpaired classes of real and fake samples, while (b) our PFA mitigates these two problems: prior and classes can be matched at the same time, only if R is accurate enough. 3.3. Pseudo-conditional Feature Adversarial Learning As discussed, we find that the main cause of the prior cha… view at source ↗

**Figure 3.** Illustration of our VER. (a) Plain regression (Narayan et al., 2020; Ye et al., 2023) need paired visual features and semantic labels. Thus, it only works on seen classes. (b) Adversarial Regression (Wang et al., 2023a) can work on both seen and unseen classes. (c) Our VER unsupervisedly pre-trains a VAE to model intra-class variations, then uses its embeddings to enhance (a) and (b) seamlessly. 3.5. Optimization a… view at source ↗

Figures from the paper (6 more)

**Figure 4.** Figure 4: The APE comparison for our I-VAEGAN and BiVAEGAN on (a) AWA1 and (b) AWA2. Our method effectively reduces APE. To evaluate how well our method mitigates the proposed APE, we calculate per-class APE from Bi-VAEGAN and our I-VAEGAN, as shown in [PITH_FULL_IMAGE:figures…

**Figure 5.** Figure 5: The comparison for prior estimation in non-uniform datasets: AWA1 and AWA2. Prior Probability 0.000 0.005 0.010 0.015 0.020 0.025 0.030 0.035 Unseen classes 0 10 20 30 40 50 Prior Probability 0.000 0.005 0.010 0.015 0.020 Unseen classes 0 10 20 30 40 50 60 70 (a) CUB C…

**Figure 6.** Figure 6: The comparison for prior estimation in close-uniform datasets: CUB and SUN. C. Additional Experiments C.1. Additional Dataset Information To demonstrate TZSL’s ability of our I-VAEGAN, we conduct experiments using four popular benchmark datasets, including two non-unif…

**Figure 7.** Figure 7: Accuracy curve on (a) AWA1 and (b) AWA2 and error curve on (c) AWA1 and (d) AWA2. 14 [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]

**Figure 8.** Figure 8: Our VER improves semantic regression across various existing methods. (a) AWA1 (b) AWA2 (c) CUB (d) SUN 𝑛𝑠𝑦𝑛 600 1200 1800 2400 3000 0.84 0.85 0.86 0.87 0.88 0.89 0.90 0.87 0.88 0.89 0.90 0.91 0.92 600 1200 1800 2400 3000 𝑛𝑠𝑦𝑛 0.64 0.66 0.68 0.70 0.72 0.74 30 60 90 120…

**Figure 9.** Figure 9: Varying the number of synthesized samples of unseen classes. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 35 canonical work pages

[1]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...
[2]

Free: Feature refinement for generalized zero-shot learning

Chen, S., Wang, W., Xia, B., Peng, Q., You, X., Zheng, F., and Shao, L. Free: Feature refinement for generalized zero-shot learning. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 122--131, 2021 a

work page 2021
[3]

Hsva: Hierarchical semantic-visual adaptation for zero-shot learning

Chen, S., Xie, G., Liu, Y., Peng, Q., Sun, B., Li, H., You, X., and Shao, L. Hsva: Hierarchical semantic-visual adaptation for zero-shot learning. Advances in Neural Information Processing Systems, 34, 2021 b

work page 2021
[4]

Transzero++: Cross attribute-guided transformer for zero-shot learning

Chen, S., Hong, Z., Hou, W., Xie, G.-S., Song, Y., Zhao, J., You, X., Yan, S., and Shao, L. Transzero++: Cross attribute-guided transformer for zero-shot learning. IEEE transactions on pattern analysis and machine intelligence, 2022 a

work page 2022
[5]

Transzero: Attribute-guided transformer for zero-shot learning

Chen, S., Hong, Z., Liu, Y., Xie, G.-s., Sun, B., Li, H., Peng, Q., Lu, K., and You, X. Transzero: Attribute-guided transformer for zero-shot learning. In AAAI, 2022 b

work page 2022
[6]

Semantics disentangling for generalized zero-shot learning

Chen, Z., Luo, Y., Qiu, R., Wang, S., Huang, Z., Li, J., and Zhang, Z. Semantics disentangling for generalized zero-shot learning. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 8712--8720, 2021 c

work page 2021
[7]

Imagenet: A large-scale hierarchical image database

Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pp.\ 248--255. Ieee, 2009

2009
[8]

M., Xiang, T., and Gong, S

Fu, Y., Hospedales, T. M., Xiang, T., and Gong, S. Transductive multi-view zero-shot learning. IEEE transactions on pattern analysis and machine intelligence, 37 0 (11): 0 2332--2345, 2015

work page 2015

Show all 45 references

[9]

Generative adversarial nets

Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. Generative adversarial nets. Advances in neural information processing systems, 27, 2014

2014
[10]

Gulrajani, I., Ahmed, F., Arjovsky, M., Dumoulin, V., and Courville, A. C. Improved training of wasserstein gans. Advances in neural information processing systems, 30, 2017

2017
[11]

Contrastive embedding for generalized zero-shot learning

Han, Z., Fu, Z., Chen, S., and Yang, J. Contrastive embedding for generalized zero-shot learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 2371--2381, 2021

2021
[12]

Deep residual learning for image recognition

He, K., Zhang, X., Ren, S., and Sun, J. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.\ 770--778, 2016

2016
[13]

H., Nickisch, H., and Harmeling, S

Lampert, C. H., Nickisch, H., and Harmeling, S. Attribute-based classification for zero-shot visual object categorization. IEEE transactions on pattern analysis and machine intelligence, 36 0 (3): 0 453--465, 2013

2013
[14]

Leveraging the invariant side of generative zero-shot learning

Li, J., Jing, M., Lu, K., Ding, Z., Zhu, L., and Huang, Z. Leveraging the invariant side of generative zero-shot learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 7402--7411, 2019 a

2019
[15]

R., and Fu, Y

Li, K., Min, M. R., and Fu, Y. Rethinking zero-shot learning: A conditional visual classification perspective. In ICCV, pp.\ 3583--3592, 2019 b

2019
[16]

Detecting and correcting for label shift with black box predictors

Lipton, Z., Wang, Y.-X., and Smola, A. Detecting and correcting for label shift with black box predictors. In International conference on machine learning, pp.\ 3122--3130. PMLR, 2018

2018
[17]

Transductive zero-shot learning with generative model-driven structure alignment

Liu, Y., Tao, K., Tian, T., Gao, X., Han, J., and Shao, L. Transductive zero-shot learning with generative model-driven structure alignment. Pattern Recognition, 153: 0 110561, 2024

2024
[18]

and Hutter, F

Loshchilov, I. and Hutter, F. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017

2017 arXiv
[19]

Transductive zero-shot learning by decoupled feature generation

Marmoreo, F., Cavazza, J., and Murino, V. Transductive zero-shot learning by decoupled feature generation. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp.\ 3109--3118, 2021

2021
[20]

Transductive zero-shot learning with reliability-based pseudo-label integration

Mohebi, S., Taheri, M., and Mansoori, E. Transductive zero-shot learning with reliability-based pseudo-label integration. In 2024 20th CSI International Symposium on Artificial Intelligence and Signal Processing (AISP), pp.\ 1--7. IEEE, 2024

2024
[21]

F., Xian, Y., Gool, L

Naeem, M. F., Xian, Y., Gool, L. V., and Tombari, F. I2dformer: Learning image to document attention for zero-shot image classification. Advances in Neural Information Processing Systems, 35: 0 12283--12294, 2022

2022
[22]

S., Snoek, C

Narayan, S., Gupta, A., Khan, F. S., Snoek, C. G., and Shao, L. Latent embedding feedback and discriminative features for zero-shot classification. In Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part XXII 16, pp.\ 479--4...

2020
[23]

Pytorch: An imperative style, high-performance deep learning library

Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al. Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems, 32, 2019

2019
[24]

and Hays, J

Patterson, G. and Hays, J. Sun attribute database: Discovering, annotating, and recognizing scene attributes. In 2012 IEEE Conference on Computer Vision and Pattern Recognition, pp.\ 2751--2758. IEEE, 2012

2012
[25]

W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al

Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pp.\ 8748--8763. PMLR, 2021

2021
[26]

Learning deep representations of fine-grained visual descriptions

Reed, S., Akata, Z., Lee, H., and Schiele, B. Learning deep representations of fine-grained visual descriptions. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.\ 49--58, 2016

2016
[27]

Generalized zero-and few-shot learning via aligned variational autoencoders

Schonfeld, E., Ebrahimi, S., Sinha, S., Darrell, T., and Akata, Z. Generalized zero-and few-shot learning via aligned variational autoencoders. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 8247--8255, 2019

2019
[28]

The caltech-ucsd birds-200-2011 dataset

Wah, C., Branson, S., Welinder, P., Perona, P., and Belongie, S. The caltech-ucsd birds-200-2011 dataset. 2011

2011
[29]

Transductive zero-shot learning with visual structure constraint

Wan, Z., Chen, D., Li, Y., Yan, X., Zhang, J., Yu, Y., and Liao, J. Transductive zero-shot learning with visual structure constraint. Advances in neural information processing systems, 32, 2019

2019
[30]

Bi-directional distribution alignment for transductive zero-shot learning

Wang, Z., Hao, Y., Mu, T., Li, O., Wang, S., and He, X. Bi-directional distribution alignment for transductive zero-shot learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 19893--19902, 2023 a

2023
[31]

Improving zero-shot generalization for clip with synthesized prompts

Wang, Z., Liang, J., He, R., Xu, N., Wang, Z., and Tan, T. Improving zero-shot generalization for clip with synthesized prompts. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pp.\ 3032--3042, 2023 b

2023
[32]

Self-supervised domain-aware generative network for generalized zero-shot learning

Wu, J., Zhang, T., Zha, Z.-J., Luo, J., Zhang, Y., and Wu, F. Self-supervised domain-aware generative network for generalized zero-shot learning. In CVPR, pp.\ 12767--12776, 2020

2020
[33]

H., Schiele, B., and Akata, Z

Xian, Y., Lampert, C. H., Schiele, B., and Akata, Z. Zero-shot learning—a comprehensive evaluation of the good, the bad and the ugly. IEEE transactions on pattern analysis and machine intelligence, 41 0 (9): 0 2251--2265, 2018 a

2018
[34]

Feature generating networks for zero-shot learning

Xian, Y., Lorenz, T., Schiele, B., and Akata, Z. Feature generating networks for zero-shot learning. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.\ 5542--5551, 2018 b

2018
[35]

f-vaegan-d2: A feature generating framework for any-shot learning

Xian, Y., Sharma, S., Schiele, B., and Akata, Z. f-vaegan-d2: A feature generating framework for any-shot learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 10275--10284, 2019

2019
[36]

Attribute prototype network for zero-shot learning

Xu, W., Xian, Y., Wang, J., Schiele, B., and Akata, Z. Attribute prototype network for zero-shot learning. In NeurIPS, 2020

2020
[37]

Attribute prototype network for any-shot learning

Xu, W., Xian, Y., Wang, J., Schiele, B., and Akata, Z. Attribute prototype network for any-shot learning. International Journal of Computer Vision, 130 0 (7): 0 1735--1753, 2022

2022
[38]

Consistency-guided pseudo labeling for transductive zero-shot learning

Yang, H., Wang, N., Wang, Z., Wang, L., and Li, H. Consistency-guided pseudo labeling for transductive zero-shot learning. Information Sciences, 670: 0 120572, 2024

2024
[39]

and Guo, Y

Ye, M. and Guo, Y. Progressive ensemble networks for zero-shot recognition. In CVPR, pp.\ 11728--11736, 2019

2019
[40]

Sr-gan: Semantic rectifying generative adversarial network for zero-shot learning

Ye, Z., Lyu, F., Li, L., Fu, Q., Ren, J., and Hu, F. Sr-gan: Semantic rectifying generative adversarial network for zero-shot learning. In 2019 IEEE international conference on multimedia and expo (ICME), pp.\ 85--90. IEEE, 2019

2019
[41]

Disentangling semantic-to-visual confusion for zero-shot learning

Ye, Z., Hu, F., Lyu, F., Li, L., and Huang, K. Disentangling semantic-to-visual confusion for zero-shot learning. IEEE Transactions on Multimedia, 2021

2021
[42]

Rebalanced zero-shot learning

Ye, Z., Yang, G., Jin, X., Liu, Y., and Huang, K. Rebalanced zero-shot learning. IEEE Transactions on Image Processing, 2023

2023
[43]

N., Jin, X., Huang, X., Xu, H., Jin, Y., and Huang, K

Ye, Z., Gowda, S. N., Jin, X., Huang, X., Xu, H., Jin, Y., and Huang, K. Exploring data efficiency in zero-shot learning with diffusion models. arXiv preprint arXiv:2406.02929, 2024

2024 arXiv
[44]

Boosting generative zero-shot learning by synthesizing diverse features with attribute augmentation

Zhao, X., Shen, Y., Wang, S., and Zhang, H. Boosting generative zero-shot learning by synthesizing diverse features with attribute augmentation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, pp.\ 3454--3462, 2022

2022
[45]

A generative adversarial approach for zero-shot learning from noisy texts

Zhu, Y., Elhoseiny, M., Liu, B., Peng, X., and Elgammal, A. A generative adversarial approach for zero-shot learning from noisy texts. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2018

2018

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

[2] [2]

Free: Feature refinement for generalized zero-shot learning

Chen, S., Wang, W., Xia, B., Peng, Q., You, X., Zheng, F., and Shao, L. Free: Feature refinement for generalized zero-shot learning. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 122--131, 2021 a

work page 2021

[3] [3]

Hsva: Hierarchical semantic-visual adaptation for zero-shot learning

Chen, S., Xie, G., Liu, Y., Peng, Q., Sun, B., Li, H., You, X., and Shao, L. Hsva: Hierarchical semantic-visual adaptation for zero-shot learning. Advances in Neural Information Processing Systems, 34, 2021 b

work page 2021

[4] [4]

Transzero++: Cross attribute-guided transformer for zero-shot learning

Chen, S., Hong, Z., Hou, W., Xie, G.-S., Song, Y., Zhao, J., You, X., Yan, S., and Shao, L. Transzero++: Cross attribute-guided transformer for zero-shot learning. IEEE transactions on pattern analysis and machine intelligence, 2022 a

work page 2022

[5] [5]

Transzero: Attribute-guided transformer for zero-shot learning

Chen, S., Hong, Z., Liu, Y., Xie, G.-s., Sun, B., Li, H., Peng, Q., Lu, K., and You, X. Transzero: Attribute-guided transformer for zero-shot learning. In AAAI, 2022 b

work page 2022

[6] [6]

Semantics disentangling for generalized zero-shot learning

Chen, Z., Luo, Y., Qiu, R., Wang, S., Huang, Z., Li, J., and Zhang, Z. Semantics disentangling for generalized zero-shot learning. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 8712--8720, 2021 c

work page 2021

[7] [7]

Imagenet: A large-scale hierarchical image database

Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pp.\ 248--255. Ieee, 2009

2009

[8] [8]

M., Xiang, T., and Gong, S

Fu, Y., Hospedales, T. M., Xiang, T., and Gong, S. Transductive multi-view zero-shot learning. IEEE transactions on pattern analysis and machine intelligence, 37 0 (11): 0 2332--2345, 2015

work page 2015