Papers with vision models
How Well Do Vision Models Encode Diagram Attributes? (2024.acl-srw)
Copied to clipboard
Haruto Yoshida, Keito Kudo, Yoichi Aoki, Ryota Tanaka, Itsumi Saito, Keisuke Sakaguchi, Kentaro Inui
| Challenge: | Experimental results show vision models struggle to identify diagram attributes such as node colors and shapes, along with edge colors and connection patterns. |
| Approach: | They evaluated vision models and retrieving diagrams using text queries to determine how well they recognize diagram attributes and edge connection patterns. |
| Outcome: | The models can recognize node colors, shapes, and edge colors, but struggle to identify differences in edge connection patterns that play a pivotal role in the semantics of diagrams. |
BiLD: Bi-directional Logits Difference Loss for Large Language Model Distillation (2025.coling-main)
Copied to clipboard
| Challenge: | Knowledge distillation (KD) is a method for reducing model size while preserving performance. |
| Approach: | They propose a method to distill large language models at the logit level by transferring knowledge from a large teacher model to a smaller student model. |
| Outcome: | The proposed method outperforms supervised fine-tuning, vanilla KL loss and five other distillation methods on 13 datasets. |
Not All Directions Matter: Towards Structured and Task-Aware Low-Rank Model Adaptation (2026.acl-long)
Copied to clipboard
Xi Xiao, Chenrui Ma, Yunbei Zhang, Chen Liu, Zhuxuanzi Wang, Yanshu Li, Lin Zhao, Guosheng Hu, Tianyang Wang, Hao Xu
| Challenge: | Low-Rank Adaptation (LoRA) is a key parameter-efficient fine-tuning method . however, its effectiveness is hampered by semantic drift and structural incoherence . |
| Approach: | They propose a low-rank Adaptation framework that tackles semantic drift and structural incoherence by pruning task-irrelevant directions. |
| Outcome: | Experiments on large language models, vision models, and vision models show that the proposed framework outperforms LoRA and advanced dynamic rank allocation and sparsity-based methods. |
CAPSTONE: Composable Attribute‐Prompted Scene Translation for Zero‐Shot Vision–Language Reasoning (2025.emnlp-industry)
Copied to clipboard
| Challenge: | CAPSTONE transforms visual inputs into structured text prompts that can be interpreted by a frozen Large Language Model (LLM). |
| Approach: | They propose a plug-and-play framework that transforms off-the-shelf vision models into structured text prompts that can be interpreted by a frozen Large Language Model (LLM). |
| Outcome: | The proposed framework outperforms fully trained VLMs on the POPE dataset while the 4B model achieves competitive results. |
VGBench: A Comprehensive Benchmark of Vector Graphics Understanding and Generation for Large Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Current vision models use pixels to rasterize the visual world, but vector graphics are not the best or unique way to represent visual content. |
| Approach: | They propose a benchmark for vector graphics processing with capable Large Language Models . they use a set of questions to evaluate vector graphics formats and a wide range of prompting techniques . |
| Outcome: | The proposed benchmark compares LLMs on rasterized representations with vector graphics . it shows that LLM models show strong capability on both aspects . |
A Computational Acquisition Model for Multimodal Word Categorization (2022.naacl-main)
Copied to clipboard
| Challenge: | Recent advances in self-supervised modeling of text and images open new opportunities for computational models of child language acquisition. |
| Approach: | They propose a multimodal language acquisition model trained from image-caption pairs on naturalistic data using cross-modal self-supervision. |
| Outcome: | The proposed model learns word categories and object recognition abilities, the authors show . their model is trained from image-caption pairs on naturalistic data using cross-modal self-supervision . |
Multi-Modal Open-Domain Dialogue (2021.emnlp-main)
Copied to clipboard
| Challenge: | Recent work in open-domain conversational agents has demonstrated that significant improvements in humanness and user preference can be achieved via massive scaling in both pre-training data and model size. |
| Approach: | They combine open-domain dialogue agents with vision models to investigate human preferences and humanness. |
| Outcome: | The proposed model outperforms existing models in multi-modal dialogue while performing as well as its predecessor (text-only) BlenderBot. |
On Vision Features in Multimodal Machine Translation (2022.acl-long)
Copied to clipboard
| Challenge: | Recent work on multimodal machine translation (MMT) has focused on the way of incorporating vision features into translation but little attention is given to the quality of vision models. |
| Approach: | They develop a selective attention model to study the patch-level contribution of an image in multimodal machine translation. |
| Outcome: | The proposed model is able to learn translation from the visual modality on probing tasks and is compared with existing models. |
Can Graph Neural Networks Learn Language with Extremely Weak Text Supervision? (2025.acl-long)
Copied to clipboard
| Challenge: | Graph Neural Networks (GNNs) with CLIP pipeline are difficult because of the scarcity of labeled data and text supervision, different levels of downstream tasks, and conceptual gaps between domains. |
| Approach: | They propose a multi-modal prompt learning paradigm to adapt pre-trained GNNs to downstream tasks with weak text supervision. |
| Outcome: | The proposed model can generalize graphs to unseen classes with weak text supervision. |
LATTE: Learning to Think with Vision Specialists (2025.emnlp-main)
Copied to clipboard
Zixian Ma, Jianguo Zhang, Zhiwei Liu, Jieyu Zhang, Juntao Tan, Manli Shu, Juan Carlos Niebles, Shelby Heinecke, Huan Wang, Caiming Xiong, Ranjay Krishna, Silvio Savarese
| Challenge: | Open-source vision-language models excel on simple question-answering tasks, but struggle with complex questions that require both perception and reasoning. |
| Approach: | They propose a family of vision-language models that have LeArned to Think wiTh vision spEcialists by offloading perception to state-of-the-art vision models. |
| Outcome: | The proposed model achieves 4-5% gains over baselines across 6 benchmarks covering both perception and reasoning abilities. |
On the Difference of BERT-style and CLIP-style Text Encoders (2023.findings-acl)
Copied to clipboard
| Challenge: | Masked language modeling is one of the most popular pretraining recipes in natural language processing. |
| Approach: | They analyze BERT-style and CLIP-style text encoders from three experiments . they show that CLIP style encoder is equipped with synesthesia for the cross-modal association . |
| Outcome: | The proposed models outperform BERT-style models on vision-centric text understanding tasks, but have synesthesia for the cross-modal association, similar to the senses of humans. |
LADDER: Language-Driven Slice Discovery and Error Rectification in Vision Classifiers (2025.findings-acl)
Copied to clipboard
Shantanu Ghosh, Rayan Syed, Chenyu Wang, Vaibhav Choudhary, Binxu Li, Clare B Poynton, Shyam Visweswaran, Kayhan Batmanghelich
| Challenge: | Current slice discovery methods in computer vision rely on converting input images into sets of attributes and testing hypotheses about configurations of pre-computed attributes associated with elevated error patterns. |
| Approach: | They propose a method to identify systematic biases in the mistakes of pre-trained vision models by converting input images into sets of attributes and testing hypotheses about configurations of these attributes. |
| Outcome: | The proposed method outperforms existing methods on 3 natural and 3 medical imaging datasets and generates pseudo-labels for each identified bias. |
On the Editability of Delta Parameters in Post-Trained Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Several studies have explored delta parameter properties via pruning, quantization, low-rank approximation, and extrapolation, but what properties of delta parameters are essential for maintaining performance? |
| Approach: | They propose to examine delta parameter properties along magnitude and sign . they propose to use a loss-based local surrogate analysis to examine editing effects . |
| Outcome: | The proposed analysis shows that delta parameters can be edited while maintaining performance. |