ModelAufgelistetExtern

GKD

> [!Note] > This repository contains model weights after domain-general distillation and task learning.

Quellehuggingface.co/yongers/GKD/tree/38a1b6e228839dec59fc29b494ef6dcaaa08a153

ArtOtherBasismodellfacebook/deit-base-patch16-224VersionRevision 38a1b6e

Modellquelle

Art: Other
Dateigröße: 332 MB–2,47 GB
Basismodell: facebook/deit-base-patch16-224
Version: vRevision 38a1b6e
Aktualisiert: 27. Sept. 2026
Downloads: 0
Aufgaben: Bildsegmentierung
Quelle: Hugging Face

Quellenauszug

[!Note] > This repository contains model weights after domain-general distillation and task learning.

Quellen

1 Quelle

Hugging FacePrimärVersion 38a1b6e228839dec59fc29b494ef6dcaaa08a153

huggingface.co/yongers/GKD/tree/38a1b6e228839dec59fc29b494ef6dcaaa08a153

Verifiziert 27. Sept.

Modellartefakte

20 Artefakte

General_distillation/DINOv2_base2small_training_deit_IMNET_QSD_epoch100_citys_epoch300_512/checkpoint_300.pth

pth · 686 MB · SHA-256 2363d5ab5907…58f2 · Hugging Face

Herunterladen

General_distillation/DINOv2_base2small_training_deit_IMNET_QSD_epoch100_GTA_epoch300_512/checkpoint_300.pth

pth · 686 MB · SHA-256 ae27464e9359…f6ab · Hugging Face

Herunterladen

General_distillation/DINOv2_base2small_training_deit_IMNET_QSD_epoch100_potsdamRGB_epoch300_512/checkpoint_300.pth

pth · 686 MB · SHA-256 8ba941dda69d…22c0 · Hugging Face

Herunterladen

General_distillation/DINOv2_base2small_training_dino_IMNET_QSD_epoch100_citys_epoch300_512/checkpoint_300.pth

pth · 686 MB · SHA-256 aec7c0224663…8fb9 · Hugging Face

Herunterladen

General_distillation/DINOv2_base2small_training_dino_IMNET_QSD_epoch100_GTA_epoch300_512/checkpoint_300.pth

pth · 686 MB · SHA-256 4d5d09d43aa6…bd4b · Hugging Face

Herunterladen

General_distillation/DINOv2_base2small_training_dino_IMNET_QSD_epoch100_potsdamRGB_epoch300_512/checkpoint_300.pth

pth · 686 MB · SHA-256 5bbb3b8703fd…9f92 · Hugging Face

Herunterladen

General_distillation/DINOv2_large2base_training_deit_IMNET_QSD_epoch100_citys_epoch300_512/checkpoint_300.pth

pth · 2,47 GB · SHA-256 5bc2043bb1a7…c0b2 · Hugging Face

Herunterladen

General_distillation/DINOv2_large2base_training_deit_IMNET_QSD_epoch100_GTA_epoch300_512/checkpoint_300.pth

pth · 2,47 GB · SHA-256 c2f5e144cad6…142a · Hugging Face

Herunterladen

General_distillation/DINOv2_large2base_training_deit_IMNET_QSD_epoch100_potsdamRGB_epoch300_512/checkpoint_300.pth

pth · 2,47 GB · SHA-256 8b7765d89f72…bb51 · Hugging Face

Herunterladen

General_distillation/DINOv2_large2base_training_dino_IMNET_QSD_epoch100_citys_epoch300_512/checkpoint_300.pth

pth · 2,47 GB · SHA-256 72018a458959…7513 · Hugging Face

Herunterladen

General_distillation/DINOv2_large2base_training_dino_IMNET_QSD_epoch100_GTA_epoch300_512/checkpoint_300.pth

pth · 2,47 GB · SHA-256 7915aa930c2f…5518 · Hugging Face

Herunterladen

General_distillation/DINOv2_large2base_training_dino_IMNET_QSD_epoch100_potsdamRGB_epoch300_512/checkpoint_300.pth

pth · 2,47 GB · SHA-256 a054bbed9bc5…1220 · Hugging Face

Herunterladen

Task_learning/DINOv2_base2small_training_deit_IMNET_QSD_epoch100_citys_epoch300_512_mask2former-40k_dg_citys-512x512.pth

pth · 332 MB · SHA-256 7020b6273349…08b5 · Hugging Face

Herunterladen

Task_learning/DINOv2_base2small_training_deit_IMNET_QSD_epoch100_GTA_epoch300_512_mask2former-40k_dg_gta5-512x512.pth

pth · 338 MB · SHA-256 590db14f31a8…7ab9 · Hugging Face

Herunterladen

Task_learning/DINOv2_base2small_training_dino_IMNET_QSD_epoch100_citys_epoch300_512_mask2former-40k_dg_citys-512x512.pth

pth · 337 MB · SHA-256 93c405655b51…73e7 · Hugging Face

Herunterladen

Task_learning/DINOv2_base2small_training_dino_IMNET_QSD_epoch100_GTA_epoch300_512_mask2former-40k_dg_gta5-512x512.pth

pth · 333 MB · SHA-256 41bb404c54b0…573d · Hugging Face

Herunterladen

Task_learning/DINOv2_large2base_training_deit_IMNET_QSD_epoch100_citys_epoch300_512_mask2former-40k_dg_citys-512x512.pth

pth · 588 MB · SHA-256 8bc84f83e507…91ff · Hugging Face

Herunterladen

Task_learning/DINOv2_large2base_training_deit_IMNET_QSD_epoch100_GTA_epoch300_512_mask2former-40k_dg_gta5-512x512.pth

pth · 586 MB · SHA-256 5cd2e2829c7d…658c · Hugging Face

Herunterladen

Task_learning/DINOv2_large2base_training_dino_IMNET_QSD_epoch100_citys_epoch300_512_mask2former-40k_dg_citys-512x512.pth

pth · 578 MB · SHA-256 e8acc1b55ff9…8bd1 · Hugging Face

Herunterladen

Task_learning/DINOv2_large2base_training_dino_IMNET_QSD_epoch100_GTA_epoch300_512_mask2former-40k_dg_gta5-512x512.pth

pth · 588 MB · SHA-256 bebd369b8fbc…c3bf · Hugging Face

Herunterladen

Quellenauszüge

2 Auszüge

Readme27. Sept.huggingface.co/yongers/GKD

--- library_name: transformers license: apache-2.0 base_model: - facebook/deit-base-patch16-224 - facebook/deit-small-patch16-224 pipeline_tag: image-segmentation --- <div align="center"> <img src="fig/logo.png" width="400px" height="400px"> </div> # [CVPR2026] Generalizable Knowledge Distillation from Vision Foundation Models for Semantic Segmentation [![arXiv](https://img.shields.io/badge/Paper-arXiv:2603.02554-Green)](https://arxiv.org/pdf/2603.02554) > [!Note] > This repository contains model weights after domain-general distillation and task learning. ## Overview - `Motivation`: Conventional knowledge distillation approaches primarily preserve in-domain accuracy while neglecting out-of-domain generalization, which is essential under distribution shifts. This limitation becomes more severe with the emergence of vision foundation models (VFMs): although VFMs exhibit strong robustness on unseen data, distilling them with conventional KD often compromises this ability. - `Methodology`: GKD decouples representation learning from task learning. In the first stage, the student acquires domain-agnostic representations through selective feature distillation, and in the second stage, these representations are frozen for task adaptation, thereby mitigating overfitting to visible domains. To further support transfer, we introduce a query-based soft distillation mechanism, where student features act as queries to teacher representations to selectively retrieve transferable spatial knowledge from VFMs. Extensive experiments on five domain generalization benchmarks demonstrate that GKD consistently outperforms existing KD methods, achieving average gains of +1.9% in foundation-to-foundation (F2F) and +10.6% in foundation-to-local (F2L) distillation. <div align="center"> <img src="fig/GKD_Fig2.png" width="100%" height="100%"> </div> ### Comparison with State-of-the-Art KD <div align="center"> <img src="fig/table_1.png" width="" height=""> </div> ### Comparison with State-of-the-Art Domain Generalization (DG) <div align="center"> <img src="fig/table_2.png" width="80%" height="80%"> </div> ---

Note27. Sept.huggingface.co/yongers/GKD

Source context: 0 downloads · 0 likes · Pipeline image-segmentation · Library transformers · Repo yongers/GKD