Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.03370

LAS-CLIP: A Lightweight Adapter Steering Approach for CLIP's Visual Encoder

Anh-Khoa Dinh-Duc·Duc-Tai Dinh·Tam V. Nguyen·Minh-Triet Tran
View original source ↗Read source document ↗Export metadata ↓

Abstract

CLIP's visual encoder produces only global image representations, limiting its use in region-level tasks. Existing adaptations rely on visual prompting, input masking, or encoder fine-tuning, each compromising pre-trained representations. We propose LAS-CLIP, a Lightweight Adapter Steering approach that keeps every CLIP parameter frozen. A compact MaskAdapter generates per-head, per-layer attention biases from an input mask and injects them into the frozen self-attention layers, steering attention toward the target region. Crucially, because the backbone remains strictly untouched, LAS-CLIP seamlessly reverts to vanilla CLIP when no mask is provided, preserving its foundational zero-shot capabilities. With approximately 116K to 145K trainable parameters and 100K training samples on two T4 GPUs, LAS-CLIP achieves competitive or superior results compared to Alpha-CLIP on ImageNet-S zero-shot classification and RefCOCO referring expression comprehension, despite the latter fine-tuning its entire encoder on millions of samples. Qualitative analysis further confirms stronger representational fidelity under incorrect masks and in downstream generation. Our project page is link to https://github.com/AnhKhoa585/lasclip

Explore related subjects

cs.CV

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Anh-Khoa Dinh-Duc, Duc-Tai Dinh, Tam V. Nguyen, Minh-Triet Tran. 2026-10-02. LAS-CLIP: A Lightweight Adapter Steering Approach for CLIP's Visual Encoder. https://arxiv.org/abs/2610.03370

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

VIDiff: Translating Videos via Multi-Modal Instructions with Diffusion Models

Diffusion models have achieved significant success in image and video generation. This motivates a growing interest in video editing tasks, where videos are edited according to provided text descriptions. However, most existing approaches only focus on video editing for short clips and rely on time-consuming tuning or inference. We are the first to propose Video Instruction Diffusion (VIDiff), a unified foundation model designed for a wide range of video tasks. These tasks encompass both understanding tasks (such as language-guided video object segmentation) and generative tasks (video editing and enhancement). Our model can edit and translate the desired results within seconds based on user instructions. Moreover, we design an iterative auto-regressive method to ensure consistency in editing and enhancing long videos. We provide convincing generative results for diverse input videos and written instructions, both qualitatively and quantitatively. More examples can be found at our website https://ChenHsing.github.io/VIDiff.

cs.CV↗
arXiv2026

Textualized and Feature-based Models for Compound Multimodal Emotion Recognition in the Wild

Systems for multimodal emotion recognition (ER) are commonly trained to extract features from different modalities (e.g., visual, audio, and textual) that are combined to predict individual basic emotions. However, compound emotions often occur in real-world scenarios, and the uncertainty of recognizing such complex emotions over diverse modalities is challenging for feature-based models. As an alternative, emerging large language models (LLMs) like BERT and LLaMA can rely on explicit non-verbal cues that may be translated from different non-textual modalities (e.g., audio and visual) into text. Textualization of modalities augments data with emotional cues to help the LLM encode the interconnections between all modalities in a shared text space. In such text-based models, prior knowledge of ER tasks is leveraged to textualize relevant non-verbal cues such as audio tone from vocal expressions, and action unit intensity from facial expressions. Since the pre-trained weights are publicly available for many LLMs, training on large-scale datasets is unnecessary, allowing to fine-tune for downstream tasks such as compound ER (CER). This paper compares the potential of text- and feature-based approaches for compound multimodal ER in videos. Experiments were conducted on the challenging C-EXPR-DB dataset in the wild for CER, and contrasted with results on the MELD dataset for basic ER. Our results indicate that multimodal textualization provides lower accuracy than feature-based models on C-EXPR-DB, where text transcripts are captured in the wild. However, higher accuracy can be achieved when the video data has rich transcripts. Our code is available.

cs.CV↗
arXiv2026

A Sobel-Gradient MLP Baseline for Handwritten Character Recognition

This study examines how much handwritten-character information is retained by a deliberately simple first-order edge representation. Instead of learning spatial filters, each input image is transformed by the fixed Sobel-Feldman operator into signed horizontal and vertical derivative maps, which are independently normalized, flattened, and classified by a multilayer perceptron (MLP). The resulting model therefore separates fixed edge extraction from learned classification and provides a controlled baseline for evaluating the sufficiency of first-order image gradients. In the executed experiments, the Sobel-gradient MLP achieves 98.54 percent test accuracy on MNIST and 92.50 percent on the TensorFlow Datasets (TFDS) EMNIST Letters configuration. Macro F1 scores are 0.9853 and 0.9265, respectively. One-vs-rest ROC analysis further yields micro/macro AUC values of 0.9998/0.9998 on MNIST and 0.9987/0.9982 on EMNIST Letters. Confusion-matrix analysis shows that the remaining errors are concentrated among geometrically similar classes, especially 3/8 and 4/9 for MNIST and I/L and G/Q for EMNIST Letters. These results show that fixed first-order gradients preserve substantial class-discriminative structure, while also revealing the specific ambiguities that remain when recognition is driven by edge geometry alone.

cs.CV↗