pith. sign in

arxiv: 2606.01947 · v1 · pith:GGX65FHQnew · submitted 2026-06-01 · 💻 cs.CV · cs.AI

Parameter-Efficient Fine-Tuning of Large Pretrained Models for Instance Segmentation Tasks

Nermeen Abou Baker , David Rohrschneider , Uwe Handmann This is my paper

Pith reviewed 2026-06-28 15:12 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords parameter-efficient fine-tuninginstance segmentationadaptersLoRAtransformerdeformable attentiontransfer learning
0
0 comments X

The pith

Sequential adapters plus LoRA on deformable attention let transformer models reach competitive instance segmentation results while updating only 1-6% of parameters.

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

The paper tests adapters and LoRA on two transformer backbones across four instance segmentation benchmarks to show that these methods can match or approach full fine-tuning accuracy. It finds that placing two or three adapters in sequence per block, combined with the first reported use of LoRA inside deformable attention, delivers the efficiency gain. A reader should care because full fine-tuning of large models normally touches 40-55% of parameters; cutting that fraction by an order of magnitude makes repeated adaptation to new domains or datasets far more practical. The work also notes that the best configuration depends on dataset difficulty and model choice, so no single recipe works everywhere.

Core claim

Integrating sequentially arranged adapter modules and applying LoRA to deformable attention achieves competitive performance on instance segmentation while fine-tuning only about 1-6% of model parameters, compared with the 40-55% required by traditional fine-tuning. Two to three adapters per transformer block give the best performance-efficiency trade-off, and LoRA on deformable attention is often more parameter-efficient than adapter configurations alone. Effectiveness varies with dataset complexity and model architecture.

What carries the argument

Sequentially arranged adapter modules combined with LoRA applied to deformable attention layers inside the transformer blocks.

If this is right

  • Two or three adapters per transformer block strike the reported optimal balance between accuracy and parameter count.
  • LoRA applied to deformable attention can exceed the efficiency of adapter-only setups on some datasets.
  • PEFT performance varies systematically with dataset complexity and model architecture.
  • Instance segmentation transfer learning becomes feasible at much lower computational cost than full fine-tuning.

Where Pith is reading between the lines

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

  • The same sequential-adapter-plus-LoRA pattern could be tested on other dense prediction tasks such as semantic segmentation or depth estimation.
  • Automated search over the number and placement of adapters might reduce the need for manual per-dataset tuning.
  • Lower fine-tuning cost could allow models to be refreshed more often when new labeled data arrives.

Load-bearing premise

The four chosen benchmark datasets and two base models are representative enough for the efficiency claims to generalize.

What would settle it

A new large-scale instance segmentation dataset where full fine-tuning reaches the same mAP while updating fewer than 6% of parameters would falsify the claimed efficiency advantage.

Figures

Figures reproduced from arXiv: 2606.01947 by David Rohrschneider, Nermeen Abou Baker, Uwe Handmann.

Figure 1
Figure 1. Figure 1: High-level view on the model architecture of Mask DINO and SEEM. 4.3. Adapters The proposed adapter architecture is inspired by the NLP-based approach introduced by Houlsby et al. [13]. In their approach, a linear layer is used to project down the output of the intermediate transformer layer, an activation function is used to introduce non-linearity at the bottleneck, a linear layer is used to project up t… view at source ↗
Figure 2
Figure 2. Figure 2: Location of adapters inside the meta-architecture (a) and detailed placement inside a self￾or cross-attention block (b + c). Added components are colored red. linear up-projection, and a residual input that is added (⊕) to the adapter output, as shown in [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: An adapter block consists of a simple feed-forward network and a residual connection (blue block). Inside one adapter (red frame), this block is sequentially repeated I times. Each of the I blocks receives the same residual input. 4.3.3. Residual connection In contrast to the original implementation, where the residual input serves as a skip connection to add up the information before the adapter module, i… view at source ↗
Figure 4
Figure 4. Figure 4: Location of LoRA inside the meta-architecture (a) and detailed placement inside a self- or cross-attention block (b + c). Added components are colored red. According to the original LoRA implementation [9], the input sequence is processed in parallel by the base weights (W0) and the LoRA weights (BA), scaled by a factor α r . Then, the results are added, leading to the following expression: h = xWT 0 + α r… view at source ↗
Figure 5
Figure 5. Figure 5: The LoRA application to regular and deformable attention mechanism. 4.4.1. Regular self- and cross-attention In the regular self- or cross-attention mechanism, Q, K and V are calculated by a matrix multiplication between the input sequence and corresponding weight matrices WQ, WK and WV. After the actual attention operation, the output is multiplied by a fourth set of weights (WO) to apply a final linear t… view at source ↗
Figure 6
Figure 6. Figure 6: Delta (δ) comparison of SEEM results [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Delta (δ) comparison of Mask DINO results (This does not include the Cityscapes results, because they were obtained with the larger Swin-L backbone). increase in the number of parameters for both models. Finetuning only the embedding weights does not capture enough information about the downstream task and results in an even larger reduction in the average delta (δ). This is exactly the problem where adapt… view at source ↗
Figure 8
Figure 8. Figure 8: Four example-visualizations (cropped and scaled) per dataset for Mask DINO, showing the ground truth, full-head, 2 adapter, and LoRA finetuning results. 6.3. Inference speed The inference times for the tested datasets and configurations, image dimensions, and the respective numbers of classes are shown in [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗
read the original abstract

Research and applications in artificial intelligence have recently shifted with the rise of large pretrained models, which deliver state-of-the-art results across numerous tasks. However, the substantial increase in parameters introduces a need for parameter-efficient training strategies. Despite significant advancements, limited research has explored parameter-efficient fine-tuning (PEFT) methods in the context of transformer-based models for instance segmentation. Addressing this gap, this study investigates the effectiveness of PEFT methods, specifically adapters and Low-Rank Adaptation (LoRA), applied to two models across four benchmark datasets. Integrating sequentially arranged adapter modules and applying LoRA to deformable attention--explored here for the first time--achieves competitive performance while fine-tuning only about 1-6% of model parameters, a marked improvement over the 40-55% required in traditional fine-tuning. Key findings indicate that using 2-3 adapters per transformer block offers an optimal balance of performance and efficiency. Furthermore, LoRA, exhibits strong parameter efficiency when applied to deformable attention, and in certain cases surpasses adapter configurations. These results show that the impact of PEFT techniques varies based on dataset complexity and model architecture, underscoring the importance of context-specific tuning. Overall, this work demonstrates the potential of PEFT to enable scalable, customizable, and computationally efficient transfer learning for instance segmentation tasks.

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 manuscript investigates parameter-efficient fine-tuning (PEFT) via adapters and LoRA (applied to deformable attention, claimed as first exploration) for transformer-based instance segmentation models. Across two base models and four benchmark datasets, it asserts that sequential adapters (optimally 2-3 per block) and LoRA achieve competitive performance while updating only 1-6% of parameters versus 40-55% for full fine-tuning, with results varying by dataset complexity and architecture.

Significance. If the empirical findings are robustly documented, the work would fill a noted gap in PEFT applications to instance segmentation and support more scalable transfer learning with large vision models by substantially reducing trainable parameters.

major comments (2)
  1. [Abstract] Abstract: the central efficiency claim (competitive performance at 1-6% parameters vs. 40-55% full fine-tuning) is stated with specific percentages but supplies no numerical results, baselines, error bars, or statistical tests, rendering the claim unverifiable from the provided text.
  2. [Abstract] Abstract / experimental claims: the assertion that 1-6% efficiency and optimality of 2-3 adapters generalize is supported only by four datasets and two models; the text itself notes variation with dataset complexity and architecture, yet no further cross-validation, additional datasets, or sensitivity analysis is described to substantiate broader applicability.
minor comments (1)
  1. [Abstract] Abstract: phrasing such as 'explored here for the first time' would benefit from a supporting citation or explicit novelty statement in the introduction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment point-by-point below and indicate planned revisions to the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central efficiency claim (competitive performance at 1-6% parameters vs. 40-55% full fine-tuning) is stated with specific percentages but supplies no numerical results, baselines, error bars, or statistical tests, rendering the claim unverifiable from the provided text.

    Authors: We acknowledge that the abstract presents the parameter-efficiency claims in summary form without embedding specific mAP values or statistical details. The full manuscript contains tables reporting exact performance metrics, parameter counts, and comparisons against full fine-tuning baselines across all datasets. We will revise the abstract to include one or two concrete quantitative highlights (e.g., mAP deltas on COCO and Cityscapes) while preserving brevity, and we will explicitly reference the experimental section for error bars and statistical comparisons. revision: yes

  2. Referee: [Abstract] Abstract / experimental claims: the assertion that 1-6% efficiency and optimality of 2-3 adapters generalize is supported only by four datasets and two models; the text itself notes variation with dataset complexity and architecture, yet no further cross-validation, additional datasets, or sensitivity analysis is described to substantiate broader applicability.

    Authors: The study deliberately evaluates two distinct transformer architectures on four standard instance-segmentation benchmarks chosen to span varying complexity. The manuscript already states that outcomes depend on dataset and architecture; we do not claim universal generalization. Adding further datasets or exhaustive sensitivity sweeps would require new large-scale experiments outside the current scope. We will expand the discussion and limitations sections to more explicitly qualify the scope of the claims and note the absence of additional cross-validation. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical comparison with no derivations

full rationale

This is an empirical study reporting experimental results of adapter and LoRA configurations on four benchmark datasets with two base models. No equations, derivations, or fitted parameters are present that could reduce to inputs by construction. Claims about 1-6% parameter efficiency and optimal 2-3 adapters are direct experimental outcomes, not self-definitional or self-citation dependent. The paper is self-contained against its own benchmarks with no load-bearing self-citations or uniqueness theorems invoked.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No mathematical derivation or theoretical model is described; the paper is an empirical benchmark study. No free parameters, axioms, or invented entities are introduced.

pith-pipeline@v0.9.1-grok · 5772 in / 1092 out tokens · 17956 ms · 2026-06-28T15:12:52.416735+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

46 extracted references · 5 canonical work pages

  1. [1]

    Segment Anything.arXiv2023, arXiv:2304.02643

    Kirillov, A.; Mintun, E.; Ravi, N.; Mao, H.; Rolland, C.; Gustafson, L.; Xiao, T.; Whitehead, S.; Berg, A.C.; Lo, W.Y.; et al. Segment Anything.arXiv2023, arXiv:2304.02643. Available online: http://arxiv.org/abs/2304.02643 (accessed on)

    Pith/arXiv arXiv

  2. [2]

    Double Transfer Learning to Detect Lithium-Ion Batteries on X-Ray Images

    Rohrschneider, D.; Abou Baker, N.; Handmann, U. Double Transfer Learning to Detect Lithium-Ion Batteries on X-Ray Images. In Proceedings of the 17th International Work-Conference on Artificial Neural Networks (IWANN), Ponta Delgada, Portugal, 19–21 June 2023; pp. 175–188

    2023

  3. [3]

    ChatGPT and Finetuned BERT: A comparative Study for Developing Intelligent Design Dupport Systems.Intell

    Qiu, Y.; Jin, Y. ChatGPT and Finetuned BERT: A comparative Study for Developing Intelligent Design Dupport Systems.Intell. Syst. Appl.2024,21, 200308. https://doi.org/10.1016/j.iswa.2023.200308

    work page doi:10.1016/j.iswa.2023.200308 2024

  4. [4]

    Few-Shot Issue Report Classification with Adapters

    Ebrahim, F.; Joy, M. Few-Shot Issue Report Classification with Adapters. In Proceedings of the International Workshop on NL-Based Software Engineering, Lisbon, Portugal, 20 April 2024; pp. 41–44

    2024

  5. [5]

    QLoRA: Efficient Finetuning of Quantized LLMs

    Dettmers, T.; Pagnoni, A.; Holtzman, A.; Zettlemoyer, L. QLoRA: Efficient Finetuning of Quantized LLMs. In Proceedings of the 37th International Conference on Neural Information Processing Systems (NeurIPS), New Orleans, LA, USA, 10–15 December 2024; pp. 10088–10115

    2024

  6. [6]

    LLaMA-Adapter: Efficient Fine-tuning of Large Language Models with Zero-initialized Attention

    Zhang, R.; Han, J.; Liu, C.; Zhou, A.; Lu, P .; Qiao, Y.; Li, H.; Gao, P . LLaMA-Adapter: Efficient Fine-tuning of Large Language Models with Zero-initialized Attention. In Proceedings of the 12th International Conference on Learning Representations (ICLR), Vienna, Austria, 7–11 May 2024

    2024

  7. [7]

    Multilingual Domain Adaptation for NMT: Decoupling Language and Domain Information with Adapters

    Stickland, A.C.; Berard, A.; Nikoulina, V . Multilingual Domain Adaptation for NMT: Decoupling Language and Domain Information with Adapters. In Proceedings of the 6th Conference on Machine Translation, Punta Cana, Dominican Republic, 10–11 November 2021; pp. 578–598. Version June 2, 2026 submitted toMach. Learn. Knowl. Extr. 24

    2021

  8. [8]

    Simple, Scalable Adaptation for Neural Machine Translation

    Bapna, A.; Firat, O. Simple, Scalable Adaptation for Neural Machine Translation. In Proceedings of the Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), Hong Kong, China, 3–7 November 2019; pp. 1538–1548

    2019

  9. [9]

    LoRA: Low-Rank Adaptation of Large Language Models

    Hu, E.J.; Shen, Y.; Wallis, P .; Allen-Zhu, Z.; Li, Y.; Wang, S.; Wang, L.; Chen, W. LoRA: Low-Rank Adaptation of Large Language Models. In Proceedings of the 10th International Conference on Learning Representations (ICLR), Online, 25–29 April 2022

    2022

  10. [10]

    Prefix-Tuning: Optimizing Continuous Prompts for Generation

    Li, X.L.; Liang, P . Prefix-Tuning: Optimizing Continuous Prompts for Generation. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, Bangkok, Thailand, 1–6 August 2021; pp. 4582–4597

    2021

  11. [11]

    Chen, G.; Liu, F.; Meng, Z.; Liang, S. Revisiting Parameter-Efficient Tuning: Are We Really There Yet? In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), Abu Dhabi, United Arab Emirates, 7–11 December 2022; pp. 2612—-2626

    2022

  12. [12]

    The Power of Scale for Parameter-Efficient Prompt Tuning

    Lester, B.; Al-Rfou, R.; Constant, N. The Power of Scale for Parameter-Efficient Prompt Tuning. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, Punta Cana, Dominican Republic, 7–11 November 2021; pp. 3045–3059

    2021

  13. [13]

    Parameter- Efficient Transfer Learning for NLP

    Houlsby, N.; Giurgiu, A.; Jastrzebski, S.; Morrone, B.; De Laroussilhe, Q.; Gesmundo, A.; Attariyan, M.; Gelly, S. Parameter- Efficient Transfer Learning for NLP . In Proceedings of the 36th International Conference on Machine Learning (ICML), Long Beach, CA, USA, 9–15 June 2019; pp. 2790–2799

    2019

  14. [14]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Dosovitskiy, A.; Beyer, L.; Kolesnikov, A.; Weissenborn, D.; Zhai, X.; Unterthiner, T.; Dehghani, M.; Minderer, M.; Heigold, G.; Gelly, S.; et al. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. In Proceedings of the 9th International Conference on Learning Representations (ICLR), Online, 3–7 May 2021

    2021

  15. [15]

    Vision Transformer Adapter for Dense Predictions

    Chen, Z.; Duan, Y.; Wang, W.; He, J.; Lu, T.; Dai, J.; Qiao, Y. Vision Transformer Adapter for Dense Predictions. In Proceedings of the 11th International Conference on Learning Representations (ICLR), Online, 1–5 May 2023

    2023

  16. [16]

    Don’t Waste SAM

    Abou Baker, N.; Handmann, U. Don’t Waste SAM. In Proceedings of the 31st European Symposium on Artificial Neural Networks (ESANN), Bruges, Belgium, 4–6 October 2023; pp. 429–434

    2023

  17. [17]

    SAM-Adapter: Adapting Segment Anything in Underperformed Scenes

    Chen, T.; Zhu, L.; Ding, C.; Cao, R.; Wang, Y.; Zhang, S.; Li, Z.; Sun, L.; Zang, Y.; Mao, P . SAM-Adapter: Adapting Segment Anything in Underperformed Scenes. In Proceedings of the International Conference on Computer Vision Workshops (ICCVW), Paris, France, 2–3 October 2023; pp. 3359–3367

    2023

  18. [18]

    Medical SAM Adapter: Adapting Segment Anything Model for Medical Image Segmentation.arXiv2023, arxiv:2304.12620

    Wu, J.; Ji, W.; Liu, Y.; Fu, H.; Xu, M.; Xu, Y.; Jin, Y. Medical SAM Adapter: Adapting Segment Anything Model for Medical Image Segmentation.arXiv2023, arxiv:2304.12620. Available online: http://arxiv.org/abs/2304.12620 (accessed on)

    arXiv

  19. [19]

    Learning Transferable Visual Models From Natural Language Supervision

    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 Proceedings of the 38th International Conference on Machine Learning (ICML), Online, 18–24 July 2021; pp. 8748–8763

    2021

  20. [20]

    CLIP-Adapter: Better Vision-Language Models with Feature Adapters.Int

    Gao, P .; Geng, S.; Zhang, R.; Ma, T.; Fang, R.; Zhang, Y.; Li, H.; Qiao, Y. CLIP-Adapter: Better Vision-Language Models with Feature Adapters.Int. J. Comput. Vis.2024,132, 581–595

    2024

  21. [21]

    Emerging Properties in Self-Supervised Vision Transformers

    Caron, M.; Touvron, H.; Misra, I.; Jégou, H.; Mairal, J.; Bojanowski, P .; Joulin, A. Emerging Properties in Self-Supervised Vision Transformers. In Proceedings of the International Conference on Computer Vision (ICCV), Online, 11–17 October 2021; pp. 9650–9660

    2021

  22. [22]

    DINOv2: Learning Robust Visual Features without Supervision.Transact

    Oquab, M.; Darcet, T.; Moutakanni, T.; Vo, H.V .; Szafraniec, M.; Khalidov, V .; Fernandez, P .; Haziza, D.; Massa, F.; El-Nouby, A.; et al. DINOv2: Learning Robust Visual Features without Supervision.Transact. Mach. Learn. Res.2024

    2024

  23. [23]

    Learning to Adapt Foundation Model DINOv2 for Capsule Endoscopy Diagnosis.arXiv2024, arxiv:2406.10508

    Zhang, B.; Chen, Y.; Bai, L.; Zhao, Y.; Sun, Y.; Yuan, Y.; Zhang, J.; Ren, H. Learning to Adapt Foundation Model DINOv2 for Capsule Endoscopy Diagnosis.arXiv2024, arxiv:2406.10508. Available online: http://arxiv.org/abs/2406.10508 (accessed on)

    arXiv

  24. [24]

    Surgical-DINO: Adapter Learning of Foundation Models for Depth Estimation in Endoscopic Surgery.Int

    Cui, B.; Islam, M.; Bai, L.; Ren, H. Surgical-DINO: Adapter Learning of Foundation Models for Depth Estimation in Endoscopic Surgery.Int. J. Comput. Assist. Radiol. Surg.2024,19, 1013–1020. https://doi.org/10.1007/s11548-024-03083-5

    work page doi:10.1007/s11548-024-03083-5 2024

  25. [25]

    Low Rank Adaptation for Stable Domain Adaptation of Vision Transformers.Opt

    Filatov, N.; Kindulov, M. Low Rank Adaptation for Stable Domain Adaptation of Vision Transformers.Opt. Mem. Neural Netw. 2023,32, 277–283. https://doi.org/10.3103/S1060992X2306005X

    work page doi:10.3103/s1060992x2306005x 2023

  26. [26]

    One-for-All: Generalized LoRA for Parameter-Efficient Fine-tuning.arXiv2023, arxiv:2306.07967

    Chavan, A.; Liu, Z.; Gupta, D.; Xing, E.; Shen, Z. One-for-All: Generalized LoRA for Parameter-Efficient Fine-tuning.arXiv2023, arxiv:2306.07967. Available online: http://arxiv.org/abs/2306.07967 (accessed on)

    arXiv

  27. [27]

    SuperLoRA: Parameter-Efficient Unified Adaptation of Multi-Layer Attention Modules.arXiv2024, arxiv:2403.11887

    Chen, X.; Liu, J.; Wang, Y.; Wang, P .P .; Brand, M.; Wang, G.; Koike-Akino, T. SuperLoRA: Parameter-Efficient Unified Adaptation of Multi-Layer Attention Modules.arXiv2024, arxiv:2403.11887. Available online: http://arxiv.org/abs/2403.11887 (accessed on)

    arXiv

  28. [28]

    Latent Diffusion Models to Enhance the Performance of Visual Defect Segmentation Networks in Steel Surface Inspection.Sensors2024,24, 6016

    Leiñena, J.; Saiz, F.A.; Barandiaran, I. Latent Diffusion Models to Enhance the Performance of Visual Defect Segmentation Networks in Steel Surface Inspection.Sensors2024,24, 6016. https://doi.org/10.3390/s24186016

    work page doi:10.3390/s24186016

  29. [29]

    Segment Everything Everywhere All at Once

    Zou, X.; Yang, J.; Zhang, H.; Li, F.; Li, L.; Wang, J.; Wang, L.; Gao, J.; Lee, Y.J. Segment Everything Everywhere All at Once. In Proceedings of the 37th International Conference on Neural Information Processing Systems (NeurIPS), Vancouver, BC, Canada, 10–15 December 2024; p. 868

    2024

  30. [30]

    Mask DINO: Towards a Unified Transformer-Based Framework for Object Detection and Segmentation

    Li, F.; Zhang, H.; Xu, H.; Liu, S.; Zhang, L.; Ni, L.M.; Shum, H.Y. Mask DINO: Towards a Unified Transformer-Based Framework for Object Detection and Segmentation. In Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR), Vancouver, BC, Canada, 18–22 June 2023; pp. 3041–3050

    2023

  31. [31]

    Scribble-Supervised Video Object Segmentation.IEEE/CAA J

    Huang, P .; Han, J.; Liu, N.; Ren, J.; Zhang, D. Scribble-Supervised Video Object Segmentation.IEEE/CAA J. Autom. Sin.2022, 9, 339–353. https://doi.org/10.1109/JAS.2021.1004210. Version June 2, 2026 submitted toMach. Learn. Knowl. Extr. 25

    work page doi:10.1109/jas.2021.1004210 2022

  32. [32]

    SAM 2: Segment Anything in Images and Videos.arXiv2024, arxiv:2408.00714

    Ravi, N.; Gabeur, V .; Hu, Y.T.; Hu, R.; Ryali, C.; Ma, T.; Khedr, H.; Rädle, R.; Rolland, C.; Gustafson, L.; et al. SAM 2: Segment Anything in Images and Videos.arXiv2024, arxiv:2408.00714. Available online: http://arxiv.org/abs/2408.00714 (accessed on)

    Pith/arXiv arXiv

  33. [33]

    Deep Residual Learning for Image Recognition

    He, K.; Zhang, X.; Ren, S.; Sun, J. Deep Residual Learning for Image Recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV , USA, 26 June –1 July 2016; pp. 770–778

    2016

  34. [34]

    Swin Transformer: Hierarchical Vision Transformer using Shifted Windows

    Liu, Z.; Lin, Y.; Cao, Y.; Hu, H.; Wei, Y.; Zhang, Z.; Lin, S.; Guo, B. Swin Transformer: Hierarchical Vision Transformer using Shifted Windows. In Proceedings of the International Conference on Computer Vision (ICCV), Online, 11–17 October 2021; pp. 9992–10002

    2021

  35. [35]

    Per-Pixel Classification is Not All You Need for Semantic Segmentation

    Cheng, B.; Schwing, A.; Kirillov, A. Per-Pixel Classification is Not All You Need for Semantic Segmentation. In Proceedings of the 35th International Conference on Neural Information Processing Systems (NeurIPS), Online, 6–14 December 2021; pp. 17864–17875

    2021

  36. [36]

    Deformable DETR: Deformable Transformers for End-to-End Object Detection

    Zhu, X.; Su, W.; Lu, L.; Li, B.; Wang, X.; Dai, J. Deformable DETR: Deformable Transformers for End-to-End Object Detection. In Proceedings of the 9th International Conference on Learning Representations (ICLR), Online, 3–7 May 2021

    2021

  37. [37]

    Focal Modulation Networks

    Yang, J.; Li, C.; Dai, X.; Gao, J. Focal Modulation Networks. In Proceedings of the 36th International Conference on Neural Information Processing Systems (NeurIPS), New Orleans, LA, USA, 28 Novemeber–9 December 2022; pp. 4203–4217

    2022

  38. [38]

    Feature Pyramid Networks for Object Detection

    Lin, T.Y.; Dollár, P .; Girshick, R.; He, K.; Hariharan, B.; Belongie, S. Feature Pyramid Networks for Object Detection. In Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 21–26 July 2017; pp. 936–944

    2017

  39. [39]

    Masked-attention Mask Transformer for Universal Image Segmenta- tion

    Cheng, B.; Misra, I.; Schwing, A.G.; Kirillov, A.; Girdhar, R. Masked-attention Mask Transformer for Universal Image Segmenta- tion. In Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, LA, USA, 19–24 June 2022; pp. 1280–1289

    2022

  40. [40]

    LoRA-Torch: PyTorch Reimplementation of LoRA

    Lin, B. LoRA-Torch: PyTorch Reimplementation of LoRA. Available online: https://github.com/Baijiong-Lin/LoRA-Torch (accessed on 14 November 2024)

    2024

  41. [41]

    NDD20: A large- Scale Few-shot Dolphin Dataset for Coarse and Fine-grained Categorisation.arXiv2020, arxiv:2005.13359

    Trotter, C.; Atkinson, G.; Sharpe, M.; Richardson, K.; McGough, A.S.; Wright, N.; Burville, B.; Berggren, P . NDD20: A large- Scale Few-shot Dolphin Dataset for Coarse and Fine-grained Categorisation.arXiv2020, arxiv:2005.13359. Available online: http://arxiv.org/abs/2005.13359 (accessed on)

    arXiv 2005

  42. [42]

    ZeroWaste Dataset: Towards Deformable Object Segmentation in Cluttered Scenes

    Bashkirova, D.; Abdelfattah, M.; Zhu, Z.; Akl, J.; Alladkani, F.; Hu, P .; Ablavsky, V .; Calli, B.; Bargal, S.A.; Saenko, K. ZeroWaste Dataset: Towards Deformable Object Segmentation in Cluttered Scenes. In Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, LA, USA, 22–24 August 2022; pp. 21147–21157

    2022

  43. [43]

    ETHSeg: An Amodel Instance Segmentation Network and a Real-world Dataset for X-Ray Waste Inspection

    Qiu, L.; Xiong, Z.; Wang, X.; Liu, K.; Li, Y.; Chen, G.; Han, X.; Cui, S. ETHSeg: An Amodel Instance Segmentation Network and a Real-world Dataset for X-Ray Waste Inspection. In Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, LA, USA, 22–24 August 2022; pp. 2273–2282

    2022

  44. [44]

    The Cityscapes Dataset for Semantic Urban Scene Understanding

    Cordts, M.; Omran, M.; Ramos, S.; Rehfeld, T.; Enzweiler, M.; Benenson, R.; Franke, U.; Roth, S.; Schiele, B. The Cityscapes Dataset for Semantic Urban Scene Understanding. In Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV , USA, 26 June–1 July 2016; pp. 3213–3223

    2016

  45. [45]

    A Simple Framework for Open-Vocabulary Segmentation and Detection

    Zhang, H.; Li, F.; Zou, X.; Liu, S.; Li, C.; Yang, J.; Zhang, L. A Simple Framework for Open-Vocabulary Segmentation and Detection. In Proceedings of the International Conference on Computer Vision (ICCV), Paris, France, 2–6 October 2023; pp. 1020–1031

    2023

  46. [46]

    OneFormer: One Transformer to Rule Universal Image Segmentation

    Jain, J.; Li, J.; Chiu, M.T.; Hassani, A.; Orlov, N.; Shi, H. OneFormer: One Transformer to Rule Universal Image Segmentation. In Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR), Vancouver, BC, Canada, 18–22 June 2023; pp. 2989–2998. Disclaimer/Publisher’s Note:The statements, opinions and data contained in all publications...

    2023