Pith. sign in

REVIEW 4 major objections 6 minor 4 cited by

ForgetMe: Evaluating Selective Forgetting in Generative Models

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read The paper introduces the ForgetMe dataset and Entangled metric as a standardized benchmark for measuring whether diffusion models truly forget a target concept.

desk verdict A useful dataset-construction pipeline undone by a self-referential, unvalidated metric; not a benchmark yet, but worth a major-revision round. read the letter →

arxiv 2504.12574 v3 pith:G7BJMSL7 submitted 2025-04-17 cs.CV

Zhenyu Yu , Mohd Yamani Inda Idris , Pei Wang This is my paper · ORCID
classification cs.CV
keywords selectiveunlearningdiffusionmodelsgenerativeevaluationmetricdatasetconstructionobjectremovalLoRAfine-tuningconcepterasure
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish a standard way to test whether a generative model has actually forgotten a specific object or concept, rather than merely learning to remove it from a given image. It introduces the ForgetMe dataset, built by an automatic, training-free pipeline that splits images into foreground, background, and mask layers, and the Entangled metric, which scores unlearning by comparing inner target regions with outer background regions. If accepted, these would give researchers and regulators a common yardstick for comparing unlearning methods across real and synthetic image categories. The paper also demonstrates the benchmark by using LoRA fine-tuning on Stable Diffusion to perform selective forgetting and reports that this approach beats established baselines on the new metric.

What carries the argument

The load-bearing object is the Entangled metric, a harmonic combination of a similarity score between inner (target) and outer (background) regions and a consistency score comparing their means and variances. This metric is used twice: first as a quality gate when building the ForgetMe dataset's background images, and second as the evaluation score for unlearned models. The dataset itself is produced by a training-free layered pipeline: SAM generates candidate segmentation masks, CLIP scores select the mask matching the target category, GPT verifies the crop, LaMa inpaints the missing background, and Stable Diffusion refines low-scoring backgrounds. An optional Side Story module merges foreground and background layers to create synthetic paired data.

What would settle it

Show a set of generated images to human raters alongside their Entangled-S scores and ask whether the target object is still present. If images the metric scores as fully unlearned are consistently recognized as still containing the object, or if images it scores as failed are judged clean, the metric is not measuring selective forgetting as claimed.

Watch

Extended reading notes

Core claim

The central claim is that selective unlearning in generative models can be evaluated with a purpose-built benchmark rather than generic image-quality metrics. Concretely, the paper claims that the Entangled metric, which combines a similarity term between the removed inner region and the preserved outer region with a consistency term on their means and variances, captures both forgetting effectiveness and background retention. It further claims that the ForgetMe dataset, constructed automatically from birds, dogs, cats, and ImageNet classes, provides a diverse testbed for this evaluation. The experiments are presented as validation: LoRA fine-tuning on background-only images yields high Entangled scores and outperforms CLIPAway and other baselines on nearly all reported metrics.

Load-bearing premise

The Entangled metric is assumed to be a valid measure of unlearning even though it is never checked against human judgment or established metrics and is used both to construct the ForgetMe dataset and to score the unlearning results.

Editorial extensions

If this is right

  • If Entangled is accepted, unlearning methods can be compared on a single scale across categories, letting model cards report a concrete forgetting score alongside image-quality metrics.
  • The ForgetMe dataset's paired original, background, and mask structure lets developers fine-tune unlearning by training on background images alone, without collecting new data.
  • Because the dataset pipeline is training-free, adding a new sensitive concept only requires running SAM, CLIP, GPT, and LaMa on new images rather than retraining a removal model.
  • Entangled-S extends evaluation to settings where the original image is unavailable, making it possible to audit deployed models using only their outputs.
  • The paper's finding that high LoRA scaling factors produce the best Entangled scores suggests that simple fine-tuning on background-only images is a viable default unlearning baseline.

Reading between the lines

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

  • Entangled's pixel-level statistics may reward visually uniform backgrounds while missing semantic reappearance, so a human-anchored or embedding-based re-identification check would be a natural complement.
  • The same automatic layering pipeline could be reused for continual unlearning: new sensitive concepts could be converted into background-only training pairs on the fly without retraining the removal framework.
  • The results leave open whether what is forgotten is the concept itself or only the context in which it appears, since training on background images shifts the conditional distribution rather than explicitly erasing an internal representation.
  • The metric's harmonic combination of similarity and consistency could transfer to other generative editing tasks, such as object removal quality assessment, where both target removal and background preservation matter.
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

4 major / 6 minor

Summary. The paper proposes ForgetMe, a dataset and evaluation framework for selective unlearning in generative models. It introduces an automatic dataset creation pipeline based on prompt-based layered editing and training-free local feature removal (SAM, LaMa, GPT, Stable Diffusion), and an Entangled evaluation metric that combines pixel-level similarity and mean/variance consistency between inner (target) and outer (background) regions, with paired (Entangled-D) and unpaired (Entangled-S) variants. The authors construct datasets from CUB-200-2011, Stanford Dogs, ImageNet, and a synthetic cat dataset, then fine-tune Stable Diffusion with LoRA to perform unlearning. They compare against CLIPAway and LoRA-based baselines, reporting lower FID/CMMD and higher Entangled and CLIP Accuracy scores for their method.

Significance. If the benchmark and metric were sound, the work would address a real gap: standardized evaluation of selective unlearning in diffusion models is missing from the literature. The proposed pipeline is fully automated and training-free, which is a practical strength, and the inclusion of both real and synthetic datasets is commendable. The idea of a paired/unpaired evaluation metric is also timely. However, the significance is severely undercut by two load-bearing problems: the evaluation metric is used to construct the very dataset on which it is later evaluated, and the metric itself has no demonstrated semantic validity. Without addressing these issues, the claimed benchmark cannot be trusted by the community.

major comments (4)
  1. [Sections 3.1.2 Step 3 and 4.3] The Entangled metric is used both to assess and refine reconstructed backgrounds during dataset construction (Section 3.1.2 Step 3) and to evaluate unlearning methods (Section 4.3). This circular self-evaluation means that the proposed baseline is trained on backgrounds that were explicitly selected to score well on the metric, so the high Entangled values in Table 2 are expected by construction. To support the benchmark claim, the authors must break this loop, for instance by constructing the dataset without Entangled-based filtering (using only GPT validation or other metrics) and then demonstrating that the relative ranking of methods is unchanged. As written, the comparison is not a fair test.
  2. [Equations (2)-(6), Table 2] The Entangled metric is purely pixel-level: S_region is an RMSE-like distance and C_inner,outer is a harmonic combination of mean and variance ratios. No term references the semantic identity of the target concept. A model that blurs the inner region or fills it with noise matched to the outer region can inflate S_inner and maintain moderate C, producing a high Entangled score while the concept remains detectable. The paper reports no human evaluation, no correlation with established perceptual or semantic metrics, and the CLIP Accuracy numbers in Table 2 are never analyzed for agreement with Entangled. Please provide a validation experiment, such as showing that a trivial blur/noise baseline scores low on Entangled, or reporting correlation between Entangled and human judgments.
  3. [Section 4.3, Table 2] The unpaired variant Entangled-S is claimed to evaluate unlearning on generated images without ground-truth masks, but the paper never specifies how the inner and outer regions are defined or located in a generated image. Table 2 reports specific Entangled-S values (e.g., 0.8119 for Ours), yet without a concrete protocol for segmenting or aligning regions in an unpaired setting, these numbers are unreproducible. The authors must either provide the exact procedure (e.g., SAM + CLIP-based localization, or fixed positional grids) or explicitly state that Entangled-S is computed on images where the mask is known; otherwise, the central unlearning evaluation is vacuous.
  4. [Table 1, Section 4.4] Table 1 reports 'Selected' and 'Success' counts with percentages (e.g., 97.44% success for Bird), but the text does not define what 'success' means. It seems related to foreground extraction, yet Section 3.1.2 Step 3 uses Entangled to filter or refine backgrounds, so the dataset construction process is intertwined with the evaluation metric. Please clarify the exact success criterion, report how many images were rejected at each stage (foreground extraction, background reconstruction), and quantify the distribution of Entangled scores during construction. Without this transparency, the dataset composition and its relationship to the metric remain unclear.
minor comments (6)
  1. [Throughout] The manuscript contains multiple unresolved placeholders, including 'Section ??' (after Eq. (2)), 'Figure ??' (Section 4.5), 'CLIPAway [?]' (Section 4.5), and 'The rationale for selecting this function is discussed in Section ??'. These must be filled in before any review.
  2. [Equation (1)] The definition of Entangled is garbled in the text: the formula appears as 'Entangled = α +β α Sinner, outer + β Cinner, outer = ...' which is not a valid mathematical expression. Please typeset the harmonic mean correctly, e.g., Entangled = (α+β) S C / (α C + β S).
  3. [Equations (2)-(3)] S_region is defined as a root-mean-square difference, which is a distance (lower is better), while S_inner,outer is defined as a harmonic combination with 1-S_outer, so the roles of 'similarity' versus 'distance' are confusing. Please clarify the direction of each component and why S_outer is flipped.
  4. [Section 4.5] The comparison states that the method 'outperforms SOTA method CLIPAway in flexibility, requiring no additional training', but CLIPAway is an object-removal method while the proposed approach performs LoRA fine-tuning. Please explain how these settings are comparable and ensure the baselines receive identical prompts, masks, and seeds.
  5. [Section 4.1-4.2] The paper promises a public release of dataset and code only upon acceptance; for a benchmark paper, this is insufficient. Please provide a public anonymized link or a commitment to release the data and code at review time to allow reproducibility.
  6. [Table 2] All metrics in Table 2 are reported as single numbers without standard deviations or multiple seeds. Given the sensitivity of LoRA training and generation, report mean±std over at least three runs.

Circularity Check

2 steps flagged · score 7.0 of 10

Entangled filters the dataset and tunes the LoRA hyperparameter, then the same Entangled scores are reported as benchmark validation; the central evidence is circular.

  1. fitted input called prediction [Section 3.1.2 'Background Reconstruction' Step 3; reused as the evaluation metric in Section 4.3 and Table 2]
    "Step 3: Evaluate background quality. The Entangled metric is used to assess the reconstructed background by measuring its residual association with the removed foreground. If the score is low, indicating poor reconstruction quality, further refinement is required. ... Entangled assesses removal by measuring similarity and consistency between inner (target) and outer (background) regions."

    Entangled is first an optimization/filter target in dataset construction: backgrounds scoring low are sent back for prompt refinement and re-inpainting (Step 4) until the metric is satisfied. The same Entangled quantity is then presented in Section 4.3 and Table 2 as an independent measure of unlearning success. A method that reproduces the pipeline's Entangled-filtered backgrounds will score high by construction, so the reported validation of the ForgetMe dataset and Entangled metric re-reads the construction criterion as evidence rather than testing it against independent ground truth.

  2. fitted input called prediction [Section 4.6 'Parameters Analysis', Table 3 and Table 2]
    "Unlearning effectiveness improved as α increased, with values above 0.9 showing significant unlearning effects while preserving structural consistency ... Table 3 shows Entangled scores for various α ... These results suggest that higher α values are preferable for effective selective unlearning while maintaining background integrity, with an α close to 1.0 recommended for rigorous tasks."

    The LoRA scaling factor α is selected by maximizing Entangled scores (Table 3), and the same Entangled-D and Entangled-S scores are then reported in Table 2 as evidence that 'Ours' is the best unlearning method. Selecting a hyperparameter against the evaluation metric and then reporting that metric as the result is fitting to the test; the advantage of the proposed framework on Entangled is at least partly manufactured by this selection loop. The paper provides no held-out split, human study, or agreement with an external semantic metric to break the loop.

full rationale

This is a benchmark/dataset paper rather than a mathematical derivation, so most circularity patterns do not apply: there is no load-bearing self-citation chain, and FID, CMMD, CLIP Distance, and CLIP Accuracy provide some external content. The core problem is that the proposed Entangled metric is used twice in the same loop: once as the filter/refinement criterion when constructing ForgetMe backgrounds (Section 3.1.2, Step 3), and again as the headline evaluation of unlearning methods (Section 4.3, Table 2). The LoRA scaling factor is also tuned against Entangled (Section 4.6). Thus the central claim that the dataset and metric are 'validated' reduces, in part, to checking that the pipeline's own construction and tuning choices score well on the pipeline's own metric. Two additional validity gaps reinforce this concern without being circular by themselves: the unpaired Entangled-S mode is never specified (no procedure is given for locating inner and outer regions in a generated image), and the consistency terms in Equations (4)-(8) are written using X (original image) statistics, so as defined they may not depend on the unlearned output at all. These issues make the Entangled-D/S values in Table 2 weak evidence of concept forgetting, but the circularity score is set at 7 rather than higher because the paper does include external metrics and does not import its central premise from a self-citation.

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

The central contribution rests on a set of weak domain assumptions about the reliability of SAM, CLIP, GPT-4o, LaMa, and Stable Diffusion, and on the Entangled metric's validity. The metric is free to vary with weights alpha/beta, and the dataset is filtered by the same pipeline, introducing circularity and selection bias. No new physical or conceptual entities are introduced.

free parameters (3)
  • Entangled metric weights alpha, beta = alpha = beta = 0.5 (default)
    Definition 3.1 introduces alpha and beta with alpha+beta=1. Changing them changes the metric's balance between similarity and consistency, and the paper gives no principled derivation for the default values.
  • LoRA scaling factor alpha = Recommended 0.9-1.0
    Section 4.6 tunes the LoRA scaling factor from 0.1 to 1.0 and reports that values above 0.9 are needed for effective unlearning. This is a free parameter fitted to the Entangled scores.
  • Dataset inclusion success thresholds = Success rates 87.54%-97.44% per dataset
    Table 1 shows that 87-97% of images are retained after SAM/CLIP/GPT/Entangled filtering. The exact thresholds for these filters are not specified, and they are effectively fitted to produce a clean dataset, which biases the benchmark.
assumptions (5)
  • domain assumption SAM segmentation masks accurately isolate the target foreground object in natural images.
    Section 3.1.1 Step 1 relies on SAM to generate masks that define the 'inner' region; if masks are inaccurate, the Entangled metric and the dataset are built on incorrect boundaries.
  • domain assumption CLIP scores correlate with semantic relevance of a mask to the target category.
    Section 3.1.1 Step 2 uses CLIP to rank masks; this assumes CLIP's embedding space is a reliable proxy for 'this region contains a dog/bird/cat'.
  • domain assumption GPT-4o validation is accurate for confirming whether a crop shows the target category.
    Section 3.1.1 Step 4 asks GPT-4o yes/no questions to accept or reject candidate foregrounds. The paper does not measure GPT-4o's error rate on this task.
  • domain assumption LaMa and Stable Diffusion inpainting can produce visually and semantically correct backgrounds where the foreground has been removed.
    Section 3.1.2 assumes these models restore structural integrity without introducing artifacts; the resulting backgrounds are treated as ground truth for the dataset.
  • domain assumption Fine-tuning on background images with LoRA causes the model to truly 'forget' the foreground concept rather than learn a conditional object-removal behavior.
    The paper frames LoRA fine-tuning as selective unlearning, but the experiments do not distinguish between genuine unlearning and simple overfitting to background-only images.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ForgetMe: Evaluating Selective Forgetting in Generative Models." pith.science (2026). https://pith.science/paper/G7BJMSL7

@misc{pith2026250412574,
  author       = {Pith},
  title        = {Pith review of: ForgetMe: Evaluating Selective Forgetting in Generative Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G7BJMSL7}},
  note         = {Machine review of arXiv:2504.12574}
}
read the original abstract

The widespread adoption of diffusion models in image generation has increased the demand for privacy-compliant unlearning. However, due to the high-dimensional nature and complex feature representations of diffusion models, achieving selective unlearning remains challenging, as existing methods struggle to remove sensitive information while preserving the consistency of non-sensitive regions. To address this, we propose an Automatic Dataset Creation Framework based on prompt-based layered editing and training-free local feature removal, constructing the ForgetMe dataset and introducing the Entangled evaluation metric. The Entangled metric quantifies unlearning effectiveness by assessing the similarity and consistency between the target and background regions and supports both paired (Entangled-D) and unpaired (Entangled-S) image data, enabling unsupervised evaluation. The ForgetMe dataset encompasses a diverse set of real and synthetic scenarios, including CUB-200-2011 (Birds), Stanford-Dogs, ImageNet, and a synthetic cat dataset. We apply LoRA fine-tuning on Stable Diffusion to achieve selective unlearning on this dataset and validate the effectiveness of both the ForgetMe dataset and the Entangled metric, establishing them as benchmarks for selective unlearning. Our work provides a scalable and adaptable solution for advancing privacy-preserving generative AI.

Figures

Figures reproduced from arXiv: 2504.12574 by the authors.

Figure 2
Figure 2. Architecture of the proposed Automatic Dataset Creation Framework, which is based on prompt-based layered editing and [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The visualization of main story and side story perspectives. The main story illustrates the layered decomposition of the original [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Comparison of model performance on selective unlearning tasks using our proposed dataset ( [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: Ablation study on the impact of α in LoRA for selective unlearning across four ForgetMe datasets: Bird, Cat, Dog, and ImageNet. α ∈ [0.1, 1.0] shows that effective unlearning is achieved only when α ≥ 0.9. 8 0.9 showing significant unlearning effects while preserv￾ing …

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Video Deepfake Abuse: How Company Choices Predictably Shape Misuse Patterns

    cs.CY 2025-11 conditional novelty 6.0 of 10

    A few open-weight video models and distribution platforms dominate the creation and spread of NSFW AI video, making developer and platform choices the main intervention points for reducing non-consensual deepfake abuse.

  2. SatelliteFormula: Multi-Modal Symbolic Regression from Remote Sensing Imagery for Physics Discovery

    cs.CV 2025-06 reject novelty 4.0 of 10

    SatelliteFormula couples a Swin Transformer image encoder with a symbolic regression decoder to generate expressions for indices such as NDVI and biomass from satellite imagery.

  3. A Multimodal Deep Learning Framework for Early Diagnosis of Liver Cancer via Optimized BiLSTM-AM-VMD Architecture

    cs.LG 2025-09 reject novelty 3.0 of 10

    The paper claims a BiLSTM-AM-VMD model achieves AUC 0.963 for early HCC diagnosis, but the evidence is undermined by contradictory dataset descriptions and missing artifacts.

  4. Multi-Modal Machine Learning Framework for Predicting Early Recurrence of Brain Tumors Using MRI and Clinical Biomarkers

    cs.LG 2025-09 reject novelty 3.0 of 10

    XGBoost combining MRI radiomics and clinical biomarkers reportedly reaches C-index 0.782 for early brain tumor recurrence, but the paper's methods describe a liver-cancer cohort and no evaluation of its claimed tempor...

Reference graph

Works this paper leans on

76 extracted references · 23 canonical work pages · cited by 4 Pith papers

  1. [1]

    On the opportunities and risks of foundation models

    Rishi Bommasani, Drew A Hudson, Ehsan Adeli, Russ Alt- man, Simran Arora, Sydney von Arx, Michael S Bernstein, Jeannette Bohg, Antoine Bosselut, Emma Brunskill, et al. On the opportunities and risks of foundation models. arXiv preprint arXiv:2108.07258, 2021. 2

    arXiv 2021

  2. [2]

    Machine unlearning in large language models

    Kongyang Chen, Zixin Wang, Bing Mi, Waixi Liu, Shaowei Wang, Xiaojun Ren, and Jiaxing Shen. Machine unlearning in large language models. arXiv preprint arXiv:2404.16841,

    arXiv

  3. [3]

    Fast model debias with machine un- learning

    Ruizhe Chen, Jianfei Yang, Huimin Xiong, Jianhong Bai, Tianxiang Hu, Jin Hao, Yang Feng, Joey Tianyi Zhou, Jian Wu, and Zuozhu Liu. Fast model debias with machine un- learning. Advances in Neural Information Processing Sys- tems, 36, 2024. 2

    2024

  4. [4]

    Score forgetting distillation: A swift, data-free method for ma- chine unlearning in diffusion models

    Tianqi Chen, Shujian Zhang, and Mingyuan Zhou. Score forgetting distillation: A swift, data-free method for ma- chine unlearning in diffusion models. arXiv preprint arXiv:2409.11219, 2024. 2

    arXiv 2024

  5. [5]

    Effectively unbiased fid and inception score and where to find them

    Min Jin Chong and David Forsyth. Effectively unbiased fid and inception score and where to find them. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6070–6079, 2020. 2, 6

    work page 2020

  6. [6]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. 3, 6

    work page 2009

  7. [7]

    Clipaway: Harmonizing focused embeddings for removing objects via diffusion models

    Yigit Ekin, Ahmet Burak Yildirim, Erdem Eren Caglar, Aykut Erdem, Erkut Erdem, and Aysegul Dundar. Clipaway: Harmonizing focused embeddings for removing objects via diffusion models. arXiv preprint arXiv:2406.09368, 2024. 6

    arXiv 2024

  8. [8]

    Salun: Empowering machine unlearn- ing via gradient-based weight saliency in both image classi- fication and generation

    Chongyu Fan, Jiancheng Liu, Yihua Zhang, Eric Wong, Den- nis Wei, and Sijia Liu. Salun: Empowering machine unlearn- ing via gradient-based weight saliency in both image classi- fication and generation. arXiv preprint arXiv:2310.12508 ,

    arXiv

Show all 76 references
  1. [9]

    Erasing concepts from diffusion models

    Rohit Gandikota, Joanna Materzynska, Jaden Fiotto- Kaufman, and David Bau. Erasing concepts from diffusion models. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 2426–2436, 2023. 2, 3

    2023

  2. [10]

    Practical unlearning for large language models

    Chongyang Gao, Lixu Wang, Chenkai Weng, Xiao Wang, and Qi Zhu. Practical unlearning for large language models. arXiv preprint arXiv:2407.10223, 2024. 3

    2024 arXiv

  3. [11]

    Meta-unlearning on diffusion models: Preventing relearning unlearned concepts

    Hongcheng Gao, Tianyu Pang, Chao Du, Taihang Hu, Zhijie Deng, and Min Lin. Meta-unlearning on diffusion models: Preventing relearning unlearned concepts. arXiv preprint arXiv:2410.12777, 2024. 2

    2024 arXiv

  4. [12]

    Ddpm-moco: Advancing industrial surface defect generation and detec- tion with generative and contrastive learning

    Yangfan He, Xinyan Wang, and Tianyu Shi. Ddpm-moco: Advancing industrial surface defect generation and detec- tion with generative and contrastive learning. In Interna- tional Joint Conference on Artificial Intelligence, pages 34–

  5. [13]

    Enhancing low-cost video editing with lightweight adaptors and temporal-aware inversion

    Yangfan He, Sida Li, Jianhui Wang, Kun Li, Xinyuan Song, Xinhang Yuan, Keqin Li, Kuan Lu, Menghao Huo, Jingqun Tang, et al. Enhancing low-cost video editing with lightweight adaptors and temporal-aware inversion. arXiv preprint arXiv:2501.04606, 2025. 2

    2025 arXiv

  6. [14]

    Selective amnesia: A contin- ual learning approach to forgetting in deep generative mod- els

    Alvin Heng and Harold Soh. Selective amnesia: A contin- ual learning approach to forgetting in deep generative mod- els. Advances in Neural Information Processing Systems, 36,

  7. [15]

    Clipscore: A reference-free evaluation met- ric for image captioning

    Jack Hessel, Ari Holtzman, Maxwell Forbes, Ronan Le Bras, and Yejin Choi. Clipscore: A reference-free evaluation met- ric for image captioning. arXiv preprint arXiv:2104.08718,

    arXiv

  8. [16]

    Lora: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021. 6

    2021 arXiv

  9. [17]

    Off- set unlearning for large language models

    James Y Huang, Wenxuan Zhou, Fei Wang, Fred Morstat- ter, Sheng Zhang, Hoifung Poon, and Muhao Chen. Off- set unlearning for large language models. arXiv preprint arXiv:2404.11045, 2024. 3

    2024 arXiv

  10. [18]

    Re- thinking fid: Towards a better evaluation metric for image generation

    Sadeep Jayasumana, Srikumar Ramalingam, Andreas Veit, Daniel Glasner, Ayan Chakrabarti, and Sanjiv Kumar. Re- thinking fid: Towards a better evaluation metric for image generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 930...

    2024

  11. [19]

    Novel dataset for fine-grained image categorization: Stanford dogs

    Aditya Khosla, Nityananda Jayadevaprakash, Bangpeng Yao, and Fei-Fei Li. Novel dataset for fine-grained image categorization: Stanford dogs. In Proc. CVPR workshop on fine-grained visual categorization (FGVC), 2011. 6

    2011

  12. [20]

    Segment any- thing

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C Berg, Wan-Yen Lo, et al. Segment any- thing. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 4015–4026, 2023. 4

    2023

  13. [21]

    Layerdiffusion: Layered controlled image editing with dif- fusion models

    Pengzhi Li, Qinxuan Huang, Yikang Ding, and Zhiheng Li. Layerdiffusion: Layered controlled image editing with dif- fusion models. In SIGGRAPH Asia 2023 Technical Commu- nications, pages 1–4. 2023. 8

    2023

  14. [22]

    Multi-modal large lan- guage model with rag strategies in soccer commentary gen- eration

    Xiang Li, Yangfan He, Shuaishuai Zu, Zhengyang Li, Tianyu Shi, Yiting Xie, and Kevin Zhang. Multi-modal large lan- guage model with rag strategies in soccer commentary gen- eration. In 2025 IEEE/CVF Winter Conference on Applica- tions of Computer Vision (WACV), pages 6197–6206. IEEE,

    2025

  15. [23]

    Fal- con: Feedback-driven adaptive long/short-term memory reinforced coding optimization system

    Zeyuan Li, Yangfan He, Lewei He, Jianhui Wang, Tianyu Shi, Bin Lei, Yuchen Li, and Qiuwu Chen. Fal- con: Feedback-driven adaptive long/short-term memory reinforced coding optimization system. arXiv preprint arXiv:2410.21349, 2024. 3

    arXiv

  16. [24]

    Cmat: A multi-agent collaboration tuning frame- work for enhancing small language models

    Xuechen Liang, Yangfan He, Meiling Tao, Yinghui Xia, Jianhui Wang, Tianyu Shi, Jun Wang, and JingSong 9 Yang. Cmat: A multi-agent collaboration tuning frame- work for enhancing small language models. arXiv preprint arXiv:2404.01663, 2024. 1

    2024 arXiv

  17. [25]

    Self-evolving agents with re- flective and memory-augmented abilities

    Xuechen Liang, Yangfan He, Yinghui Xia, Xinyuan Song, Jianhui Wang, Meiling Tao, Li Sun, Xinhang Yuan, Ji- ayi Su, Keqin Li, et al. Self-evolving agents with re- flective and memory-augmented abilities. arXiv preprint arXiv:2409.00872, 2024. 3

    2024 arXiv

  18. [26]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceeding...

    2014

  19. [27]

    Towards safer large language models through machine unlearning

    Zheyuan Liu, Guangyao Dou, Zhaoxuan Tan, Yijun Tian, and Meng Jiang. Towards safer large language models through machine unlearning. arXiv preprint arXiv:2402.10058, 2024. 2

    2024 arXiv

  20. [28]

    A survey on feder- ated unlearning: Challenges, methods, and future directions

    Ziyao Liu, Yu Jiang, Jiyuan Shen, Minyi Peng, Kwok-Yan Lam, Xingliang Yuan, and Xiaoning Liu. A survey on feder- ated unlearning: Challenges, methods, and future directions. ACM Computing Surveys, 57(1):1–38, 2024. 2

    2024

  21. [29]

    Mace: Mass concept erasure in diffu- sion models

    Shilin Lu, Zilan Wang, Leyang Li, Yanzhu Liu, and Adams Wai-Kin Kong. Mace: Mass concept erasure in diffu- sion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 6430– 6440, 2024. 2

    2024

  22. [30]

    Learning and unlearning: New challenges for teach- ing in conservatoires

    Erica McWilliam, Gemma Carey, Paul Draper, and Don Lebler. Learning and unlearning: New challenges for teach- ing in conservatoires. Australian journal of music education, (1):25–31, 2006. 2

    2006

  23. [31]

    Lumina- image 2.0: A unified and efficient image generative frame- work

    Qi Qin, Le Zhuo, Yi Xin, Ruoyi Du, Zhen Li, Bin Fu, Yiting Lu, Jiakang Yuan, Xinyue Li, Dongyang Liu, et al. Lumina- image 2.0: A unified and efficient image generative frame- work. arXiv preprint arXiv:2503.21758, 2025. 1

    2025 arXiv

  24. [32]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 4

    2022

  25. [33]

    Rord: A real-world object removal dataset

    Min-Cheol Sagong, Yoon-Jae Yeo, Seung-Won Jung, and Sung-Jea Ko. Rord: A real-world object removal dataset. In BMVC, page 542, 2022. 3

    2022

  26. [34]

    Reasoning in computer vision: Taxonomy, models, tasks, and methodologies

    Ayushman Sarkar, Mohd Yamani Idna Idris, and Zhenyu Yu. Reasoning in computer vision: Taxonomy, models, tasks, and methodologies. arXiv preprint arXiv:2508.10523, 2025. 3

    2025

  27. [35]

    Unstable unlearning: The hidden risk of concept resurgence in diffusion models

    Vinith M Suriyakumar, Rohan Alur, Ayush Sekhari, Manish Raghavan, and Ashia C Wilson. Unstable unlearning: The hidden risk of concept resurgence in diffusion models. arXiv preprint arXiv:2410.08074, 2024. 2

    2024

  28. [36]

    Resolution-robust large mask inpainting with fourier convolutions

    Roman Suvorov, Elizaveta Logacheva, Anton Mashikhin, Anastasia Remizova, Arsenii Ashukha, Aleksei Silvestrov, Naejin Kong, Harshith Goka, Kiwoong Park, and Victor Lempitsky. Resolution-robust large mask inpainting with fourier convolutions. InProceedings of the IEEE/CVF winter...

    2022

  29. [37]

    To forget or not? towards practical knowl- edge unlearning for large language models

    Bozhong Tian, Xiaozhuan Liang, Siyuan Cheng, Qingbin Liu, Mengru Wang, Dianbo Sui, Xi Chen, Huajun Chen, and Ningyu Zhang. To forget or not? towards practical knowl- edge unlearning for large language models. arXiv preprint arXiv:2407.01920, 2024. 3

    2024 arXiv

  30. [38]

    The caltech-ucsd birds-200-2011 dataset

    Catherine Wah, Steve Branson, Peter Welinder, Pietro Per- ona, and Serge Belongie. The caltech-ucsd birds-200-2011 dataset. 2011. 6

    2011

  31. [39]

    Evaluation and analysis of hal- lucination in large vision-language models

    Junyang Wang, Yiyang Zhou, Guohai Xu, Pengcheng Shi, Chenlin Zhao, Haiyang Xu, Qinghao Ye, Ming Yan, Ji Zhang, Jihua Zhu, et al. Evaluation and analysis of hal- lucination in large vision-language models. arXiv preprint arXiv:2308.15126, 2023. 1

    2023 arXiv

  32. [40]

    Enhancing code llms with reinforce- ment learning in code generation: A survey

    Junqiao Wang, Zeng Zhang, Yangfan He, Zihao Zhang, Yuyang Song, Tianyu Shi, Yuchen Li, Hengyuan Xu, Kunyu Wu, Xin Yi, et al. Enhancing code llms with reinforce- ment learning in code generation: A survey. arXiv preprint arXiv:2412.20367, 2024. 1

    2024 arXiv

  33. [41]

    Mdanet: A multi-stage domain adaptation framework for generaliz- able low-light image enhancement

    Jianhui Wang, Yangfan He, Kun Li, Sida Li, Lan Zhao, Jun Yin, Miao Zhang, Tianyu Shi, and Xueqian Wang. Mdanet: A multi-stage domain adaptation framework for generaliz- able low-light image enhancement. Neurocomputing, 627: 129572, 2025. 1

    2025

  34. [42]

    Twin co-adaptive dialogue for progres- sive image generation

    Jianhui Wang, Yangfan He, Yan Zhong, Xinyuan Song, Ji- ayi Su, Yuheng Feng, Hongyang He, Wenyu Zhu, Xinhang Yuan, Kuan Lu, et al. Twin co-adaptive dialogue for progres- sive image generation. arXiv preprint arXiv:2504.14868 ,

  35. [43]

    See the forest and the trees: A synergistic reasoning framework for knowledge-based vi- sual question answering

    Junjie Wang, Yunhan Tang, Yijie Wang, Zhihao Yuan, Huan Wang, Yangfan He, and Bin Li. See the forest and the trees: A synergistic reasoning framework for knowledge-based vi- sual question answering. arXiv preprint arXiv:2507.17659,

    arXiv

  36. [44]

    Unitmge: Uniform text-motion generation and editing model via diffusion

    Ruoyu Wang, Yangfan He, Tengjiao Sun, Xiang Li, and Tianyu Shi. Unitmge: Uniform text-motion generation and editing model via diffusion. In 2025 IEEE/CVF Winter Con- ference on Applications of Computer Vision (WACV), pages 6104–6114. IEEE, 2025. 3

    2025

  37. [45]

    Evaluating deep unlearning in large lan- guage models

    Ruihan Wu, Chhavi Yadav, Russ Salakhutdinov, and Kama- lika Chaudhuri. Evaluating deep unlearning in large lan- guage models. arXiv preprint arXiv:2410.15153, 2024. 2

    2024

  38. [46]

    Regrap-llava: Reasoning enabled graph-based personalized large language and vision assistant

    Yifan Xiang, Zhenxi Zhang, Bin Li, Yixuan Weng, Shoujun Zhou, Yangfan He, and Keqin Li. Regrap-llava: Reasoning enabled graph-based personalized large language and vision assistant. arXiv preprint arXiv:2505.03654, 2025. 2

    2025 arXiv

  39. [47]

    Self-training with label-feature-consistency for do- main adaptation

    Yi Xin, Siqi Luo, Pengsheng Jin, Yuntao Du, and Chongjun Wang. Self-training with label-feature-consistency for do- main adaptation. In International Conference on Database Systems for Advanced Applications , pages 84–99. Springer,

  40. [48]

    Vmt-adapter: Parameter-efficient transfer learning for multi- task dense scene understanding

    Yi Xin, Junlong Du, Qiang Wang, Zhiwen Lin, and Ke Yan. Vmt-adapter: Parameter-efficient transfer learning for multi- task dense scene understanding. In Proceedings of the AAAI Conference on Artificial Intelligence , pages 16085–16093,

  41. [49]

    Mmap: Multi-modal alignment prompt for cross- 10 domain multi-task learning

    Yi Xin, Junlong Du, Qiang Wang, Ke Yan, and Shouhong Ding. Mmap: Multi-modal alignment prompt for cross- 10 domain multi-task learning. InProceedings of the AAAI Con- ference on Artificial Intelligence, pages 16076–16084, 2024. 1

    2024

  42. [50]

    V-petl bench: A unified visual parameter- efficient transfer learning benchmark

    Yi Xin, Siqi Luo, Xuyang Liu, Haodi Zhou, Xinyu Cheng, Christina E Lee, Junlong Du, Haozhe Wang, MingCai Chen, Ting Liu, et al. V-petl bench: A unified visual parameter- efficient transfer learning benchmark. Advances in Neural Information Processing Systems, 37:80522–80535, 2024. 1

    2024

  43. [51]

    Parameter-efficient fine-tuning for pre-trained vision models: A survey

    Yi Xin, Siqi Luo, Haodi Zhou, Junlong Du, Xiaohong Liu, Yue Fan, Qing Li, and Yuntao Du. Parameter-efficient fine-tuning for pre-trained vision models: A survey. arXiv preprint arXiv:2402.02242, 2024

    2024

  44. [52]

    Lumina-mgpt 2.0: Stand- alone autoregressive image modeling

    Yi Xin, Juncheng Yan, Qi Qin, Zhen Li, Dongyang Liu, Shicheng Li, Victor Shea-Jay Huang, Yupeng Zhou, Ren- rui Zhang, Le Zhuo, et al. Lumina-mgpt 2.0: Stand- alone autoregressive image modeling. arXiv preprint arXiv:2507.17801, 2025. 1

    2025 arXiv

  45. [53]

    Resurrect mask autoregressive modeling for efficient and scalable image generation

    Yi Xin, Le Zhuo, Qi Qin, Siqi Luo, Yuewen Cao, Bin Fu, Yangfan He, Hongsheng Li, Guangtao Zhai, Xiaohong Liu, et al. Resurrect mask autoregressive modeling for efficient and scalable image generation. arXiv preprint arXiv:2507.13032, 2025. 1

    2025 arXiv

  46. [54]

    Machine unlearning: Solutions and challenges

    Jie Xu, Zihan Wu, Cong Wang, and Xiaohua Jia. Machine unlearning: Solutions and challenges. IEEE Transactions on Emerging Topics in Computational Intelligence, 2024. 3

    2024

  47. [55]

    Machine unlearning for traditional models and large language models: A short survey

    Yi Xu. Machine unlearning for traditional models and large language models: A short survey. arXiv preprint arXiv:2404.01206, 2024. 2

    2024 arXiv

  48. [56]

    Wcdt: World-centric diffusion transformer for traffic scene generation

    Chen Yang, Yangfan He, Aaron Xuxiang Tian, Dong Chen, Jianhui Wang, Tianyu Shi, Arsalan Heydarian, and Pei Liu. Wcdt: World-centric diffusion transformer for traffic scene generation. arXiv preprint arXiv:2404.02082, 2024. 1

    2024 arXiv

  49. [57]

    Large language model unlearning

    Yuanshun Yao, Xiaojun Xu, and Yang Liu. Large language model unlearning. arXiv preprint arXiv:2310.10683, 2023. 2

    2023 arXiv

  50. [58]

    mplug-owl: Modularization empowers large language models with multimodality

    Qinghao Ye, Haiyang Xu, Guohai Xu, Jiabo Ye, Ming Yan, Yiyang Zhou, Junyang Wang, Anwen Hu, Pengcheng Shi, Yaya Shi, et al. mplug-owl: Modularization empowers large language models with multimodality. arXiv preprint arXiv:2304.14178, 2023. 1

    2023 arXiv

  51. [59]

    Towards understanding the working mechanism of text-to-image dif- fusion model

    Mingyang Yi, Aoxue Li, Yi Xin, and Zhenguo Li. Towards understanding the working mechanism of text-to-image dif- fusion model. Advances in Neural Information Processing Systems, 37:55342–55369, 2024. 1

    2024

  52. [60]

    Score: Story coherence and retrieval enhancement for ai narratives

    Qiang Yi, Yangfan He, Jianhui Wang, Xinyuan Song, Shiyao Qian, Xinhang Yuan, Li Sun, Yi Xin, Jingqun Tang, Keqin Li, et al. Score: Story coherence and retrieval enhancement for ai narratives. arXiv preprint arXiv:2503.23512, 2025. 1

    2025

  53. [61]

    Inst-inpaint: Instructing to remove objects with diffusion models

    Ahmet Burak Yildirim, Vedat Baday, Erkut Erdem, Aykut Erdem, and Aysegul Dundar. Inst-inpaint: Instructing to remove objects with diffusion models. arXiv preprint arXiv:2304.03246, 2023. 2, 3, 6

    2023 arXiv

  54. [62]

    Ai for science: A comprehensive review on in- novations, challenges, and future directions

    Zhenyu Yu. Ai for science: A comprehensive review on in- novations, challenges, and future directions. International Journal of Artificial Intelligence for Science (IJAI4S) , 1(1),

  55. [63]

    Yuan: Yielding unblem- ished aesthetics through a unified network for visual im- perfections removal in generated images

    Zhenyu Yu and Chee Seng Chan. Yuan: Yielding unblem- ished aesthetics through a unified network for visual im- perfections removal in generated images. arXiv preprint arXiv:2501.08505, 2025. 2

    2025 arXiv

  56. [64]

    From physics to foundation models: A review of ai-driven quantitative remote sensing inversion

    Zhenyu Yu, Mohd Yamani Idna Idris, Hua Wang, Pei Wang, Junyi Chen, and Kun Wang. From physics to foundation models: A review of ai-driven quantitative remote sensing inversion. arXiv preprint arXiv:2507.09081, 2025. 3

    2025 arXiv

  57. [65]

    Dc4cr: When cloud removal meets diffusion control in remote sens- ing

    Zhenyu Yu, Mohd Yamani Idna Idris, and Pei Wang. Dc4cr: When cloud removal meets diffusion control in remote sens- ing. arXiv preprint arXiv:2504.14785, 2025. 2

    2025 arXiv

  58. [66]

    Physics-constrained symbolic regression from imagery

    Zhenyu Yu, MOHD Y AMANI IDNA IDRIS, and Pei Wang. Physics-constrained symbolic regression from imagery. In 2nd AI for Math Workshop@ ICML 2025, 2025. 2

    2025

  59. [67]

    Dancetext: Point-driven interactive text and im- age layer editing using diffusion models

    Zhenyu Yu, Mohd Yamani Idna Idris, Pei Wang, and Yue- long Xia. Dancetext: Point-driven interactive text and im- age layer editing using diffusion models. arXiv preprint arXiv:2504.14108, 2025. 2

    2025

  60. [68]

    Qrs-trs: Style transfer-based image-to- image translation for carbon stock estimation in quantitative remote sensing

    Zhenyu Yu, Jinnian Wang, Hanqing Chen, and Mohd Ya- mani Idna Idris. Qrs-trs: Style transfer-based image-to- image translation for carbon stock estimation in quantitative remote sensing. IEEE Access, 2025. 2

    2025

  61. [69]

    Transparent image layer diffusion using latent transparency

    Lvmin Zhang and Maneesh Agrawala. Transparent image layer diffusion using latent transparency. arXiv preprint arXiv:2402.17113, 2024. 8

    2024 arXiv

  62. [70]

    Analyzing and mitigating object hallucination in large vision-language models

    Yiyang Zhou, Chenhang Cui, Jaehong Yoon, Linjun Zhang, Zhun Deng, Chelsea Finn, Mohit Bansal, and Huaxiu Yao. Analyzing and mitigating object hallucination in large vision-language models. arXiv preprint arXiv:2310.00754,

    arXiv

  63. [71]

    Calibrated self-rewarding vision language models

    Yiyang Zhou, Zhiyuan Fan, Dongjie Cheng, Sihan Yang, Zhaorun Chen, Chenhang Cui, Xiyao Wang, Yun Li, Linjun Zhang, and Huaxiu Yao. Calibrated self-rewarding vision language models. Advances in Neural Information Process- ing Systems, 37:51503–51531, 2024. 1

    2024

  64. [72]

    Reagent-v: A reward-driven multi-agent framework for video understanding

    Yiyang Zhou, Yangfan He, Yaofeng Su, Siwei Han, Joel Jang, Gedas Bertasius, Mohit Bansal, and Huaxiu Yao. Reagent-v: A reward-driven multi-agent framework for video understanding. arXiv preprint arXiv:2506.01300 ,

    arXiv

  65. [73]

    Glimpse: Do large vision-language models truly think with videos or just glimpse at them? arXiv preprint arXiv:2507.09491, 2025

    Yiyang Zhou, Linjie Li, Shi Qiu, Zhengyuan Yang, Yuyang Zhao, Siwei Han, Yangfan He, Kangqi Li, Haonian Ji, Zihao Zhao, et al. Glimpse: Do large vision-language models truly think with videos or just glimpse at them? arXiv preprint arXiv:2507.09491, 2025. 1

    2025 arXiv

  66. [74]

    Anyprefer: An agentic framework for preference data synthesis

    Yiyang Zhou, Zhaoyang Wang, Tianle Wang, Shangyu Xing, Peng Xia, Bo Li, Kaiyuan Zheng, Zijian Zhang, Zhaorun Chen, Wenhao Zheng, et al. Anyprefer: An agentic framework for preference data synthesis. arXiv preprint arXiv:2504.19276, 2025. 1

    2025 arXiv

  67. [75]

    Human-centric reward optimization for reinforcement learning-based auto- mated driving using large language models

    Ziqi Zhou, Jingyue Zhang, Jingyuan Zhang, Yangfan He, Boyue Wang, Tianyu Shi, and Alaa Khamis. Human-centric reward optimization for reinforcement learning-based auto- mated driving using large language models. arXiv preprint arXiv:2405.04135, 2024. 1

    2024 arXiv

  68. [76]

    Choose your anchor wisely: Effective unlearning diffusion 11 models via concept reconditioning

    Jingyu Zhu, Ruiqi Zhang, Licong Lin, and Song Mei. Choose your anchor wisely: Effective unlearning diffusion 11 models via concept reconditioning. In Neurips Safe Genera- tive AI Workshop 2024. 2 12

    2024

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.