pith. sign in

arxiv: 1907.00612 · v1 · pith:YI7SG3YUnew · submitted 2019-07-01 · 💻 cs.CV · cs.LG

One Network for Multi-Domains: Domain Adaptive Hashing with Intersectant Generative Adversarial Network

Tao He , Yuan-Fang Li , Lianli Gao , Dongxiang Zhang , Jingkuan Song This is my paper

Pith reviewed 2026-05-25 12:09 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords domain adaptive hashinggenerative adversarial networksimage retrievalcross-domain learningsemantic common spacedomain adaptationhash codesmulti-domain hashing
0
0 comments X

The pith

Encoding images from two domains into a shared semantic space and using two independent GANs for crosswise reconstruction produces aligned hash codes that work across domains.

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

The paper develops an end-to-end system for learning hash codes that remain effective when images come from a source domain and a target domain with large distributional differences. It maps both sets of images into one semantic common space. Two separate generative adversarial networks then reconstruct images from each domain using the representation of the other domain. This crosswise reconstruction is intended to shrink domain disparity and improve alignment so the same hash function supports accurate recognition and retrieval on both domains. Experiments on four public benchmarks show gains over prior methods in object recognition and image retrieval tasks.

Core claim

Our method encodes two domains images into a semantic common space, followed by two independent generative adversarial networks arming at crosswise reconstructing two domains' images, reducing domain disparity and improving alignment in the shared space. We evaluate our framework on four public benchmark datasets, all of which show that our method is superior to the other state-of-the-art methods on the tasks of object recognition and image retrieval.

What carries the argument

The intersectant generative adversarial network formed by two independent GANs that perform crosswise image reconstruction between domains.

If this is right

  • A single hash function can be learned and applied directly to both source and target domains.
  • The reconstruction objective enforces semantic consistency across domains inside the shared space.
  • Hash codes become usable for multi-domain image retrieval without separate adaptation steps.
  • End-to-end training integrates hashing, classification, and cross-domain alignment in one model.
  • The approach applies to any pair of domains that exhibit distributional shift.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same reconstruction idea could be extended to three or more domains by adding additional GANs.
  • Performance might be tested on modalities beyond images, such as video or text, where domain shifts occur.
  • If the shared space truly captures semantics, the codes could support retrieval to entirely unseen domains not seen during training.

Load-bearing premise

Independent crosswise reconstruction by the two GANs alone is sufficient to align the domains and yield discriminative hash codes without extra constraints on the shared space or explicit domain-invariant feature learning.

What would settle it

On the four benchmark datasets, if the method produces lower mean average precision for retrieval or lower accuracy for recognition than existing domain-adaptive hashing approaches, the claim of improved alignment and superiority would be refuted.

Figures

Figures reproduced from arXiv: 1907.00612 by Dongxiang Zhang, Jingkuan Song, Lianli Gao, Tao He, Yuan-Fang Li.

Figure 1
Figure 1. Figure 1: Our model consists of three components: (1) one [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 1
Figure 1. Figure 1: The overview of our framework, which is consisted of five networks: encoder, two independent generators and two distinct [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: A visualization of the target domain features learned with [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
read the original abstract

With the recent explosive increase of digital data, image recognition and retrieval become a critical practical application. Hashing is an effective solution to this problem, due to its low storage requirement and high query speed. However, most of past works focus on hashing in a single (source) domain. Thus, the learned hash function may not adapt well in a new (target) domain that has a large distributional difference with the source domain. In this paper, we explore an end-to-end domain adaptive learning framework that simultaneously and precisely generates discriminative hash codes and classifies target domain images. Our method encodes two domains images into a semantic common space, followed by two independent generative adversarial networks arming at crosswise reconstructing two domains' images, reducing domain disparity and improving alignment in the shared space. We evaluate our framework on {four} public benchmark datasets, all of which show that our method is superior to the other state-of-the-art methods on the tasks of object recognition and image retrieval.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes an end-to-end domain-adaptive hashing framework that maps source and target domain images into a shared semantic space via a common encoder, then employs two independent GANs for crosswise image reconstruction to reduce domain disparity and produce aligned, discriminative hash codes for object recognition and retrieval. It asserts superiority over state-of-the-art methods on four public benchmarks.

Significance. If the empirical superiority claims hold with proper validation, the approach would offer a generative mechanism for implicit domain alignment in hashing without explicit feature-level adversarial losses, which could be useful for cross-domain retrieval. No machine-checked proofs, reproducible code, or parameter-free derivations are present to strengthen the contribution.

major comments (2)
  1. [Abstract] Abstract: the assertion that the method is 'superior to the other state-of-the-art methods' on four benchmarks is unsupported by any quantitative results, tables, baselines, error bars, or ablation details, making the central empirical claim unverifiable from the manuscript.
  2. [Proposed Method] Method description (intersectant GAN architecture): the claim that two independent crosswise reconstruction GANs suffice to 'reduc[e] domain disparity and improv[e] alignment in the shared space' lacks any explicit regularizer on the encoder output (e.g., adversarial feature matching, hash-code cycle consistency, or domain-invariant loss); reconstruction can succeed without latent alignment, leaving the alignment mechanism unverified.
minor comments (1)
  1. [Abstract] Abstract contains placeholder braces around '{four}' and the nonstandard phrasing 'arming at' (should be 'aiming at').

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major comment below and indicate planned revisions to the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the assertion that the method is 'superior to the other state-of-the-art methods' on four benchmarks is unsupported by any quantitative results, tables, baselines, error bars, or ablation details, making the central empirical claim unverifiable from the manuscript.

    Authors: The full manuscript contains an experimental section with quantitative comparisons on the four benchmarks, including tables with baselines and metrics for recognition and retrieval. We agree, however, that the abstract itself does not reference these results. We will revise the abstract to include brief quantitative highlights of the observed improvements. revision: yes

  2. Referee: [Proposed Method] Method description (intersectant GAN architecture): the claim that two independent crosswise reconstruction GANs suffice to 'reduc[e] domain disparity and improv[e] alignment in the shared space' lacks any explicit regularizer on the encoder output (e.g., adversarial feature matching, hash-code cycle consistency, or domain-invariant loss); reconstruction can succeed without latent alignment, leaving the alignment mechanism unverified.

    Authors: The shared encoder produces hash codes that are directly fed into the crosswise GAN generators. Successful reconstruction of the opposite domain therefore requires the codes to capture semantics that are invariant across domains; otherwise the generated images would not match the target distribution. This bidirectional reconstruction supplies the alignment pressure without an additional explicit term on the encoder. We will expand the method section with a paragraph clarifying this implicit mechanism and its relation to cycle-consistency ideas. revision: partial

Circularity Check

0 steps flagged

No derivation chain present; architecture proposal only

full rationale

The paper describes a neural architecture (shared encoder + two independent cross-domain GANs) for domain-adaptive hashing. No equations, fitted parameters, predictions, or uniqueness theorems appear in the provided text. Claims rest on the empirical performance of the proposed design rather than any reduction of outputs to inputs by construction. Self-citations, if present in the full manuscript, are not load-bearing for any derivation. This is the normal case of an engineering contribution whose validity is tested externally.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The framework rests on standard assumptions of GAN training stability and the existence of a semantic common space; no explicit free parameters or invented physical entities are named in the abstract.

axioms (1)
  • domain assumption A shared semantic space exists that supports both discriminative hashing and cross-domain image reconstruction.
    Invoked when the method encodes both domains into one space before applying the GANs.
invented entities (1)
  • Intersectant Generative Adversarial Network no independent evidence
    purpose: Perform crosswise reconstruction between source and target domains to reduce disparity.
    New component introduced to arm the hashing process; no independent evidence outside the proposed framework is provided.

pith-pipeline@v0.9.0 · 5715 in / 1214 out tokens · 30889 ms · 2026-05-25T12:09:39.376740+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages · 2 internal anchors

  1. [1]

    Carreira-Perpi˜n´an and Ramin Raziperchikolaei

    [Carreira-Perpi˜n´an and Raziperchikolaei, 2015] Miguel ´A. Carreira-Perpi˜n´an and Ramin Raziperchikolaei. Hashing with binary autoencoders. In CVPR, pages 557–566,

    work page 2015

  2. [2]

    Denker, W

    [Denker et al., 1989] John S. Denker, W. R. Gardner, Hans Peter Graf, Donnie Henderson, R. E. Howard, W. Hubbard, L. D. Jackel, Henry S. Baird, and Isabelle Guyon. Neural network recognizer for hand-written zip code digits. In D. S. Touretzky, editor, NIPS, pages 323–

    work page 1989

  3. [3]

    Learning to hash with binary deep neural network

    [Do et al., 2016] Thanh-Toan Do, Anh-Dzung Doan, and Ngai-Man Cheung. Learning to hash with binary deep neural network. In ECCV, pages 219–234,

    work page 2016

  4. [4]

    Lempitsky

    [Ganin and Lempitsky, 2015] Yaroslav Ganin and Victor S. Lempitsky. Unsupervised domain adaptation by backprop- agation. In ICML, pages 1180–1189,

    work page 2015

  5. [5]

    Deep Reconstruction-Classification Networks for Unsupervised Domain Adaptation

    [Ghifary et al., 2016] Muhammad Ghifary, W. Bastiaan Kleijn, Mengjie Zhang, David Balduzzi, and Wen Li. Deep reconstruction-classification networks for unsuper- vised domain adaptation. CoRR, abs/1607.03516,

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  6. [6]

    Iterative quantization: A procrustean approach to learning binary codes for large- scale image retrieval

    [Gong et al., 2013] Yunchao Gong, Svetlana Lazebnik, Al- bert Gordo, and Florent Perronnin. Iterative quantization: A procrustean approach to learning binary codes for large- scale image retrieval. IEEE Trans. Pattern Anal. Mach. Intell., 35(12):2916–2929,

    work page 2013

  7. [7]

    Generative adversarial nets

    [Goodfellow et al., 2014] Ian Goodfellow, Jean Pouget- Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, and Sherjil Ozair. Generative adversarial nets. In NIPS, pages 2672–2680. Curran Associates, Inc.,

    work page 2014

  8. [8]

    Efros, and Trevor Darrell

    [Hoffman et al., 2018] Judy Hoffman, Eric Tzeng, Taesung Park, Jun-Yan Zhu, Phillip Isola, Kate Saenko, Alexei A. Efros, and Trevor Darrell. Cycada: Cycle-consistent ad- versarial domain adaptation. In ICML, pages 1994–2003,

    work page 2018

  9. [9]

    Duplex generative adversarial network for unsupervised domain adaptation

    [Hu et al., 2018] Lanqing Hu, Meina Kan, Shiguang Shan, and Xilin Chen. Duplex generative adversarial network for unsupervised domain adaptation. In CVPR, June

    work page 2018

  10. [10]

    Gradient-based learning ap- plied to document recognition

    [LeCun et al., 1998] Yann LeCun, L ´eon Bottou, Yoshua Bengio, Patrick Haffner, et al. Gradient-based learning ap- plied to document recognition. Proceedings of the IEEE , 86(11):2278–2324,

    work page 1998

  11. [11]

    Self-supervised adversar- ial hashing networks for cross-modal retrieval

    [Li et al., 2018] Chao Li, Cheng Deng, Ning Li, Wei Liu, Xinbo Gao, and Dacheng Tao. Self-supervised adversar- ial hashing networks for cross-modal retrieval. In CVPR, June

    work page 2018

  12. [12]

    Cou- pled generative adversarial networks

    [Liu and Tuzel, 2016] Ming-Yu Liu and Oncel Tuzel. Cou- pled generative adversarial networks. In NIPS, pages 469– 477,

    work page 2016

  13. [13]

    [Long et al., 2015] Mingsheng Long, Yue Cao, Jianmin Wang, and Michael I. Jordan. Learning transferable fea- tures with deep adaptation networks. In ICML, pages 97– 105,

    work page 2015

  14. [14]

    Kriegman, Ravi Ramamoorthi, and Kyungnam Kim

    [Murez et al., 2018] Zak Murez, Soheil Kolouri, David J. Kriegman, Ravi Ramamoorthi, and Kyungnam Kim. Im- age to image translation for domain adaptation. In CVPR, pages 4500–4509,

    work page 2018

  15. [15]

    Reading digits in natural images with unsupervised feature learn- ing

    [Netzer et al., 2011] Yuval Netzer, Tao Wang, Adam Coates, Alessandro Bissacco, Bo Wu, and Andrew Y Ng. Reading digits in natural images with unsupervised feature learn- ing. In NIPS workshop on deep learning and unsupervised feature learning, volume 2011, page 5,

    work page 2011

  16. [16]

    Adapting visual category models to new domains

    [Saenko et al., 2010] Kate Saenko, Brian Kulis, Mario Fritz, and Trevor Darrell. Adapting visual category models to new domains. In ECCV, pages 213–226. Springer,

    work page 2010

  17. [17]

    Castillo, and Rama Chellappa

    [Sankaranarayanan et al., 2018] Swami Sankaranarayanan, Yogesh Balaji, Carlos D. Castillo, and Rama Chellappa. Generate to adapt: Aligning domains using generative ad- versarial networks. In CVPR, pages 8503–8512,

    work page 2018

  18. [18]

    Deep Domain Confusion: Maximizing for Domain Invariance

    [Tzeng et al., 2014] Eric Tzeng, Judy Hoffman, Ning Zhang, Kate Saenko, and Trevor Darrell. Deep domain confusion: Maximizing for domain invariance.CoRR, abs/1412.3474,

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  19. [19]

    Adversarial discriminative domain adaptation

    [Tzeng et al., 2017] Eric Tzeng, Judy Hoffman, Kate Saenko, and Trevor Darrell. Adversarial discriminative domain adaptation. In CVPR, pages 2962–2971,

    work page 2017

  20. [20]

    Deep hashing network for unsupervised do- main adaptation

    [Venkateswara et al., 2017] Hemanth Venkateswara, Jose Eusebio, Shayok Chakraborty, and Sethuraman Pan- chanathan. Deep hashing network for unsupervised do- main adaptation. In CVPR, pages 5385–5394,

    work page 2017

  21. [21]

    Learning semantic representations for unsu- pervised domain adaptation

    [Xie et al., 2018] Shaoan Xie, Zibin Zheng, Liang Chen, and Chuan Chen. Learning semantic representations for unsu- pervised domain adaptation. In ICML, pages 5419–5428,

    work page 2018

  22. [22]

    Collaborative and adversarial network for unsupervised domain adaptation

    [Zhang et al., 2018] Weichen Zhang, Wanli Ouyang, Wen Li, and Dong Xu. Collaborative and adversarial network for unsupervised domain adaptation. In CVPR, June 2018

    work page 2018