SD‑LoRA introduces a training‑time structural distillation framework that embeds token‑level graph reasoning into a lightweight LoRA‑adapted CLIP model for few‑shot vision‑language adaptation. During training, a heterogeneous graph teacher captures patch‑patch, global‑local, and visual‑text relations over high‑resolution ViT tokens, and its structural bias is transferred to the CLIP‑LoRA backbone via implicit distillation and Prototype Predictive Alignment. At inference, the graph teacher is discarded, requiring only a single CLIP‑LoRA forward pass and a cache‑based classifier, yielding significant accuracy gains across 11 few‑shot benchmarks without extra computation.
arXiv:2609.10321v1 Announce Type: new Abstract: Knowledge distillation offers an efficient route to transfer a task-adapted vision-language teacher to a compact student. The training target in curren...
arXiv:2607. 17467v1 Announce Type: cross Abstract: Few-shot Test-Time Domain Adaptation (FSTT-DA) seeks to adapt models to novel domains using only a handful of unlabeled target samples.
arXiv:2505.07675v3 Announce Type: replace-cross Abstract: Semi-supervised learning (SSL) has emerged as a practical solution for addressing data scarcity challenges by leveraging unlabeled data. Rece...
arXiv:2605.10247v2 Announce Type: replace Abstract: Applying Large Language Models (LLMs) to graph-structured data usually involves multi-step pipelines in which textual node attributes are compresse...
arXiv:2602. 15727v2 Announce Type: replace-cross Abstract: Visual analogy learning enables image editing via demonstration rather than textual description, allowing users to specify complex transformations difficult to articulate in words.
MLLMCLIP introduces a heterogeneous distillation framework that transfers multimodal knowledge from a generative Multimodal Large Language Model (MLLM) teacher directly into a discriminative CLIP student, eliminating the need for synthetic hard negatives. The method uses an attention-based per-layer token selection and a CKA-based distillation loss to bridge architectural differences between the two models. As a result, MLLMCLIP achieves state‑of‑the‑art compositional accuracy and improves zero‑shot classification and image‑text retrieval performance.
arXiv:2607. 28967v1 Announce Type: cross Abstract: Prompt tuning adapts vision--language models with few trainable parameters, but existing approaches trade off efficiency and adaptation: static textual prompts can overfit source classes, image-conditioned prompts add per-instance computation, and multimodal tuning modifies the visual branch.
arXiv:2507.00754v3 Announce Type: replace Abstract: The integration of Large Language Model (LLMs) blocks with Vision Transformers (ViTs) holds immense promise for vision-only tasks by leveraging the...
The paper introduces UINO-FSS, a unified framework for few-shot semantic segmentation that merges knowledge from DINOv2 and SAM into a single-encoder architecture. It employs coarse-to-fine multimodal distillation, a bottleneck adapter for embedding alignment, a meta-visual prompt generator, and a mask decoder, enhanced by Mamba-based 4D correlation mining. Experiments on PASCAL-5i and COCO-20i demonstrate state‑of‑the‑art performance in 1‑shot settings, achieving 80.6% mIoU on PASCAL-5i and 64.5% on COCO-20i.
Pretrained vision-language models such as CLIP excel at zero-shot recognition but often fail at compositionality, particularly attribute-object and relational structures. Recent studies mitigate this...
arXiv:2607. 19128v1 Announce Type: new Abstract: Vision-language models (VLMs) provide a unified representation space for textual and visual information, yet their potential as general-purpose backbones for graph-structured data remains largely unexplored.
The paper investigates few-shot adaptation of vision-language models, noting that gradient magnitudes are poorly calibrated while the signs of a few top-magnitude gradients remain informative. It proposes two state‑free optimizers, SignSparseSGD and DynSignSparseSGD, which use gradient magnitudes only to select update locations and fixed‑magnitude signs to determine update directions. Across 11 datasets, three CLIP backbones, and five shot levels, these methods match or surpass parameter‑efficient and optimization‑based baselines, show reduced sensitivity to long training schedules, and require no task‑specific modules or optimizer state beyond a single shared configuration.