PyTorchKR Papers

최신 논문

총 398편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2410.01534

Toward a Holistic Evaluation of Robustness in CLIP Models

Contrastive Language-Image Pre-training (CLIP) models have shown significant potential, particularly in zero-shot classification across diverse distribution shifts. Building on existing evaluations of overall classification robustness, this work aims to provide a more comprehensive assessment of CLIP by introducing several new perspectives. First, we investigate their robustness to variations in specific visual factors. Second, we assess two critical safety objectives--confidence uncertainty and out-of-distribution detection--beyond mere classification accuracy. Third, we evaluate the finesse with which CLIP models bridge the image and text modalities. Fourth, we extend our examination to 3D awareness in CLIP models, moving beyond traditional 2D image understanding. Finally, we explore the interaction between vision and language encoders within modern large multimodal models (LMMs) that utilize CLIP as the visual backbone, focusing on how this interaction impacts classification robustness. In each aspect, we consider the impact of six factors on CLIP models: model architecture, training distribution, training set size, fine-tuning, contrastive loss, and test-time prompts. Our study uncovers several previously unknown insights into CLIP. For instance, the architecture of the visual encoder in CLIP plays a significant role in their robustness against 3D corruption. CLIP models tend to exhibit a bias towards shape when making predictions. Moreover, this bias tends to diminish after fine-tuning on ImageNet. Vision-language models like LLaVA, leveraging the CLIP vision encoder, could exhibit benefits in classification performance for challenging categories over CLIP alone. Our findings are poised to offer valuable guidance for enhancing the robustness and reliability of CLIP models.

arXiv:2410.01407공식 코드 · Papers with Code

AgriCLIP: Adapting CLIP for Agriculture and Livestock via Domain-Specialized Cross-Model Alignment

Capitalizing on vast amount of image-text data, large-scale vision-language pre-training has demonstrated remarkable zero-shot capabilities and has been utilized in several applications. However, models trained on general everyday web-crawled data often exhibit sub-optimal performance for specialized domains, likely due to domain shift. Recent works have tackled this problem for some domains (e.g., healthcare) by constructing domain-specialized image-text data. However, constructing a dedicated large-scale image-text dataset for sustainable area of agriculture and livestock is still open to research. Further, this domain desires fine-grained feature learning due to the subtle nature of the downstream tasks (e.g, nutrient deficiency detection, livestock breed classification). To address this we present AgriCLIP, a vision-language foundational model dedicated to the domain of agriculture and livestock. First, we propose a large-scale dataset, named ALive, that leverages customized prompt generation strategy to overcome the scarcity of expert annotations. Our ALive dataset covers crops, livestock, and fishery, with around 600,000 image-text pairs. Second, we propose a training pipeline that integrates both contrastive and self-supervised learning to learn both global semantic and local fine-grained domain-specialized features. Experiments on diverse set of 20 downstream tasks demonstrate the effectiveness of AgriCLIP framework, achieving an absolute gain of 7.8\% in terms of average zero-shot classification accuracy, over the standard CLIP adaptation via domain-specialized ALive dataset. Our ALive dataset and code can be accessible at \href{https://github.com/umair1221/AgriCLIP/tree/main}{Github}.

arXiv:2410.00182

Zero-Shot Classification of Crisis Tweets Using Instruction-Finetuned Large Language Models

Social media posts are frequently identified as a valuable source of open-source intelligence for disaster response, and pre-LLM NLP techniques have been evaluated on datasets of crisis tweets. We assess three commercial large language models (OpenAI GPT-4o, Gemini 1.5-flash-001 and Anthropic Claude-3-5 Sonnet) capabilities in zero-shot classification of short social media posts. In one prompt, the models are asked to perform two classification tasks: 1) identify if the post is informative in a humanitarian context; and 2) rank and provide probabilities for the post in relation to 16 possible humanitarian classes. The posts being classified are from the consolidated crisis tweet dataset, CrisisBench. Results are evaluated using macro, weighted, and binary F1-scores. The informative classification task, generally performed better without extra information, while for the humanitarian label classification providing the event that occurred during which the tweet was mined, resulted in better performance. Further, we found that the models have significantly varying performance by dataset, which raises questions about dataset quality.

arXiv:2409.19425

From Unimodal to Multimodal: Scaling up Projectors to Align Modalities

Recent contrastive multimodal vision-language models like CLIP have demonstrated robust open-world semantic understanding, becoming the standard image backbones for vision-language applications due to their aligned latent space. However, this practice has left powerful unimodal encoders for both vision and language underutilized in multimodal applications which raises a key question: Is there a plausible way to connect unimodal backbones for zero-shot vision-language tasks? To this end, we propose a novel approach that aligns vision and language modalities using only projection layers on pretrained, frozen unimodal encoders. Our method exploits the high semantic similarity between embedding spaces of well-trained vision and language models. It involves selecting semantically similar encoders in the latent space, curating a concept-rich dataset of image-caption pairs, and training simple MLP projectors. We evaluated our approach on 12 zero-shot classification datasets and 2 image-text retrieval datasets. Our best model, utilizing DINOv2 and All-Roberta-Large text encoder, achieves 76\(\%\) accuracy on ImageNet with a 20-fold reduction in data and 65 fold reduction in compute requirements. The proposed framework enhances the accessibility of model development while enabling flexible adaptation across diverse scenarios, offering an efficient approach to building multimodal models by utilizing existing unimodal architectures. Code and datasets will be released soon.

arXiv:2409.17156

An Art-centric perspective on AI-based content moderation of nudity

At a time when the influence of generative Artificial Intelligence on visual arts is a highly debated topic, we raise the attention towards a more subtle phenomenon: the algorithmic censorship of artistic nudity online. We analyze the performance of three "Not-Safe-For-Work'' image classifiers on artistic nudity, and empirically uncover the existence of a gender and a stylistic bias, as well as evident technical limitations, especially when only considering visual information. Hence, we propose a multi-modal zero-shot classification approach that improves artistic nudity classification. From our research, we draw several implications that we hope will inform future research on this topic.

arXiv:2409.17601

CleanerCLIP: Fine-grained Counterfactual Semantic Augmentation for Backdoor Defense in Contrastive Learning

Pre-trained large models for multimodal contrastive learning, such as CLIP, have been widely recognized in the industry as highly susceptible to data-poisoned backdoor attacks. This poses significant risks to downstream model training. In response to such potential threats, finetuning offers a simpler and more efficient defense choice compared to retraining large models with augmented data. In the supervised learning domain, fine-tuning defense strategies can achieve excellent defense performance. However, in the unsupervised and semi-supervised domain, we find that when CLIP faces some complex attack techniques, the existing fine-tuning defense strategy, CleanCLIP, has some limitations on defense performance. The synonym substitution of its text-augmentation is insufficient to enhance the text feature space. To compensate for this weakness, we improve it by proposing a fine-grained \textbf{T}ext \textbf{A}lignment \textbf{C}leaner (TA-Cleaner) to cut off feature connections of backdoor triggers. We randomly select a few samples for positive and negative subtext generation at each epoch of CleanCLIP, and align the subtexts to the images to strengthen the text self-supervision. We evaluate the effectiveness of our TA-Cleaner against six attack algorithms and conduct comprehensive zero-shot classification tests on ImageNet1K. Our experimental results demonstrate that TA-Cleaner achieves state-of-the-art defensiveness among finetuning-based defense techniques. Even when faced with the novel attack technique BadCLIP, our TA-Cleaner outperforms CleanCLIP by reducing the ASR of Top-1 and Top-10 by 52.02\% and 63.88\%, respectively.

arXiv:2409.09741

Benchmarking LLMs in Political Content Text-Annotation: Proof-of-Concept with Toxicity and Incivility Data

This article benchmarked the ability of OpenAI's GPTs and a number of open-source LLMs to perform annotation tasks on political content. We used a novel protest event dataset comprising more than three million digital interactions and created a gold standard that includes ground-truth labels annotated by human coders about toxicity and incivility on social media. We included in our benchmark Google's Perspective algorithm, which, along with GPTs, was employed throughout their respective APIs while the open-source LLMs were deployed locally. The findings show that Perspective API using a laxer threshold, GPT-4o, and Nous Hermes 2 Mixtral outperform other LLM's zero-shot classification annotations. In addition, Nous Hermes 2 and Mistral OpenOrca, with a smaller number of parameters, are able to perform the task with high performance, being attractive options that could offer good trade-offs between performance, implementing costs and computing time. Ancillary findings using experiments setting different temperature levels show that although GPTs tend to show not only excellent computing time but also overall good levels of reliability, only open-source LLMs ensure full reproducibility in the annotation.

arXiv:2409.07480

EEG-Language Modeling for Pathology Detection

Multimodal language modeling has enabled breakthroughs for representation learning, yet remains unexplored in the realm of functional brain data for pathology detection. This paper pioneers EEG-language models (ELMs) trained on clinical reports and 15000 EEGs. We propose to combine multimodal alignment in this novel domain with timeseries cropping and text segmentation, enabling an extension based on multiple instance learning to alleviate misalignment between irrelevant EEG or text segments. Our multimodal models significantly improve pathology detection compared to EEG-only models across four evaluations and for the first time enable zero-shot classification as well as retrieval of both neural signals and reports. In sum, these results highlight the potential of ELMs, representing significant progress for clinical applications.

arXiv:2409.00408

Multi-label Zero-Shot Audio Classification with Temporal Attention

Zero-shot learning models are capable of classifying new classes by transferring knowledge from the seen classes using auxiliary information. While most of the existing zero-shot learning methods focused on single-label classification tasks, the present study introduces a method to perform multi-label zero-shot audio classification. To address the challenge of classifying multi-label sounds while generalizing to unseen classes, we adapt temporal attention. The temporal attention mechanism assigns importance weights to different audio segments based on their acoustic and semantic compatibility, thus enabling the model to capture the varying dominance of different sound classes within an audio sample by focusing on the segments most relevant for each class. This leads to more accurate multi-label zero-shot classification than methods employing temporally aggregated acoustic features without weighting, which treat all audio segments equally. We evaluate our approach on a subset of AudioSet against a zero-shot model using uniformly aggregated acoustic features, a zero-rule baseline, and the proposed method in the supervised scenario. Our results show that temporal attention enhances the zero-shot audio classification performance in multi-label scenario.

arXiv:2409.07048

Pushing the Limits of Vision-Language Models in Remote Sensing without Human Annotations

The prominence of generalized foundation models in vision-language integration has witnessed a surge, given their multifarious applications. Within the natural domain, the procurement of vision-language datasets to construct these foundation models is facilitated by their abundant availability and the ease of web crawling. Conversely, in the remote sensing domain, although vision-language datasets exist, their volume is suboptimal for constructing robust foundation models. This study introduces an approach to curate vision-language datasets by employing an image decoding machine learning model, negating the need for human-annotated labels. Utilizing this methodology, we amassed approximately 9.6 million vision-language paired datasets in VHR imagery. The resultant model outperformed counterparts that did not leverage publicly available vision-language datasets, particularly in downstream tasks such as zero-shot classification, semantic localization, and image-text retrieval. Moreover, in tasks exclusively employing vision encoders, such as linear probing and k-NN classification, our model demonstrated superior efficacy compared to those relying on domain-specific vision-language datasets.

arXiv:2409.03938

Deep Clustering of Remote Sensing Scenes through Heterogeneous Transfer Learning

This paper proposes a method for unsupervised whole-image clustering of a target dataset of remote sensing scenes with no labels. The method consists of three main steps: (1) finetuning a pretrained deep neural network (DINOv2) on a labelled source remote sensing imagery dataset and using it to extract a feature vector from each image in the target dataset, (2) reducing the dimension of these deep features via manifold projection into a low-dimensional Euclidean space, and (3) clustering the embedded features using a Bayesian nonparametric technique to infer the number and membership of clusters simultaneously. The method takes advantage of heterogeneous transfer learning to cluster unseen data with different feature and label distributions. We demonstrate the performance of this approach outperforming state-of-the-art zero-shot classification methods on several remote sensing scene classification datasets.

arXiv:2409.03521

Have Large Vision-Language Models Mastered Art History?

The emergence of large Vision-Language Models (VLMs) has recently established new baselines in image classification across multiple domains. However, the performance of VLMs in the specific task of artwork classification, particularly art style classification of paintings - a domain traditionally mastered by art historians - has not been explored yet. Artworks pose a unique challenge compared to natural images due to their inherently complex and diverse structures, characterized by variable compositions and styles. Art historians have long studied the unique aspects of artworks, with style prediction being a crucial component of their discipline. This paper investigates whether large VLMs, which integrate visual and textual data, can effectively predict the art historical attributes of paintings. We conduct an in-depth analysis of four VLMs, namely CLIP, LLaVA, OpenFlamingo, and GPT-4o, focusing on zero-shot classification of art style, author and time period using two public benchmarks of artworks. Additionally, we present ArTest, a well-curated test set of artworks, including pivotal paintings studied by art historians.

arXiv:2409.00698공식 코드 · Papers with Code

Enhancing Remote Sensing Vision-Language Models for Zero-Shot Scene Classification

Vision-Language Models for remote sensing have shown promising uses thanks to their extensive pretraining. However, their conventional usage in zero-shot scene classification methods still involves dividing large images into patches and making independent predictions, i.e., inductive inference, thereby limiting their effectiveness by ignoring valuable contextual information. Our approach tackles this issue by utilizing initial predictions based on text prompting and patch affinity relationships from the image encoder to enhance zero-shot capabilities through transductive inference, all without the need for supervision and at a minor computational cost. Experiments on 10 remote sensing datasets with state-of-the-art Vision-Language Models demonstrate significant accuracy improvements over inductive zero-shot classification. Our source code is publicly available on Github: https://github.com/elkhouryk/RS-TransCLIP

arXiv:2409.01936공식 코드 · Papers with Code

Optimizing CLIP Models for Image Retrieval with Maintained Joint-Embedding Alignment

Contrastive Language and Image Pairing (CLIP), a transformative method in multimedia retrieval, typically trains two neural networks concurrently to generate joint embeddings for text and image pairs. However, when applied directly, these models often struggle to differentiate between visually distinct images that have similar captions, resulting in suboptimal performance for image-based similarity searches. This paper addresses the challenge of optimizing CLIP models for various image-based similarity search scenarios, while maintaining their effectiveness in text-based search tasks such as text-to-image retrieval and zero-shot classification. We propose and evaluate two novel methods aimed at refining the retrieval capabilities of CLIP without compromising the alignment between text and image embeddings. The first method involves a sequential fine-tuning process: initially optimizing the image encoder for more precise image retrieval and subsequently realigning the text encoder to these optimized image embeddings. The second approach integrates pseudo-captions during the retrieval-optimization phase to foster direct alignment within the embedding space. Through comprehensive experiments, we demonstrate that these methods enhance CLIP's performance on various benchmarks, including image retrieval, k-NN classification, and zero-shot text-based classification, while maintaining robustness in text-to-image retrieval. Our optimized models permit maintaining a single embedding per image, significantly simplifying the infrastructure needed for large-scale multi-modal similarity search systems.

arXiv:2409.01883공식 코드 · Papers with Code

Boosting Vision-Language Models for Histopathology Classification: Predict all at once

The development of vision-language models (VLMs) for histo-pathology has shown promising new usages and zero-shot performances. However, current approaches, which decompose large slides into smaller patches, focus solely on inductive classification, i.e., prediction for each patch is made independently of the other patches in the target test data. We extend the capability of these large models by introducing a transductive approach. By using text-based predictions and affinity relationships among patches, our approach leverages the strong zero-shot capabilities of these new VLMs without any additional labels. Our experiments cover four histopathology datasets and five different VLMs. Operating solely in the embedding space (i.e., in a black-box setting), our approach is highly efficient, processing $10^5$ patches in just a few seconds, and shows significant accuracy improvements over inductive zero-shot classification. Code available at https://github.com/FereshteShakeri/Histo-TransCLIP.

arXiv:2409.02136공식 코드 · Papers with Code

Large Language Models versus Classical Machine Learning: Performance in COVID-19 Mortality Prediction Using High-Dimensional Tabular Data

Background: This study aimed to evaluate and compare the performance of classical machine learning models (CMLs) and large language models (LLMs) in predicting mortality associated with COVID-19 by utilizing a high-dimensional tabular dataset. Materials and Methods: We analyzed data from 9,134 COVID-19 patients collected across four hospitals. Seven CML models, including XGBoost and random forest (RF), were trained and evaluated. The structured data was converted into text for zero-shot classification by eight LLMs, including GPT-4 and Mistral-7b. Additionally, Mistral-7b was fine-tuned using the QLoRA approach to enhance its predictive capabilities. Results: Among the CML models, XGBoost and RF achieved the highest accuracy, with F1 scores of 0.87 for internal validation and 0.83 for external validation. In the LLM category, GPT-4 was the top performer with an F1 score of 0.43. Fine-tuning Mistral-7b significantly improved its recall from 1% to 79%, resulting in an F1 score of 0.74, which was stable during external validation. Conclusion: While LLMs show moderate performance in zero-shot classification, fine-tuning can significantly enhance their effectiveness, potentially aligning them closer to CML models. However, CMLs still outperform LLMs in high-dimensional tabular data tasks.

arXiv:2409.01466

PoliPrompt: A High-Performance Cost-Effective LLM-Based Text Classification Framework for Political Science

Recent advancements in large language models (LLMs) have opened new avenues for enhancing text classification efficiency in political science, surpassing traditional machine learning methods that often require extensive feature engineering, human labeling, and task-specific training. However, their effectiveness in achieving high classification accuracy remains questionable. This paper introduces a three-stage in-context learning approach that leverages LLMs to improve classification accuracy while minimizing experimental costs. Our method incorporates automatic enhanced prompt generation, adaptive exemplar selection, and a consensus mechanism that resolves discrepancies between two weaker LLMs, refined by an advanced LLM. We validate our approach using datasets from the BBC news reports, Kavanaugh Supreme Court confirmation, and 2018 election campaign ads. The results show significant improvements in classification F1 score (+0.36 for zero-shot classification) with manageable economic costs (-78% compared with human labeling), demonstrating that our method effectively addresses the limitations of traditional machine learning while offering a scalable and reliable solution for text analysis in political science.

arXiv:2408.15802

Visual Prompt Engineering for Medical Vision Language Models in Radiology

Medical image classification in radiology faces significant challenges, particularly in generalizing to unseen pathologies. In contrast, CLIP offers a promising solution by leveraging multimodal learning to improve zero-shot classification performance. However, in the medical domain, lesions can be small and might not be well represented in the embedding space. Therefore, in this paper, we explore the potential of visual prompt engineering to enhance the capabilities of Vision Language Models (VLMs) in radiology. Leveraging BiomedCLIP, trained on extensive biomedical image-text pairs, we investigate the impact of embedding visual markers directly within radiological images to guide the model's attention to critical regions. Our evaluation on the JSRT dataset, focusing on lung nodule malignancy classification, demonstrates that incorporating visual prompts $\unicode{x2013}$ such as arrows, circles, and contours $\unicode{x2013}$ significantly improves classification metrics including AUROC, AUPRC, F1 score, and accuracy. Moreover, the study provides attention maps, showcasing enhanced model interpretability and focus on clinically relevant areas. These findings underscore the efficacy of visual prompt engineering as a straightforward yet powerful approach to advance VLM performance in medical image analysis.

arXiv:2408.13320공식 코드 · Papers with Code

Online Zero-Shot Classification with CLIP

Vision-language pre-training such as CLIP enables zero-shot transfer that can classify images according to the candidate class names. While CLIP demonstrates an impressive zero-shot performance on diverse downstream tasks, the distribution from the target data has not been leveraged sufficiently. In this work, we study a novel online zero-shot transfer scenario, where each image arrives in a random order for classification and is visited only once to obtain prediction immediately without storing its representation. Compared with the vanilla zero-shot classification, the proposed framework preserves its flexibility for online service while considering the statistics of the arrived images as the side information to capture the distribution of target data, which can help improve the performance of real-world applications. To tackle the challenge of effective online optimization, we first develop online label learning to model the target data distribution. Then, the proxy of each class in the vision space is further optimized with the proposed online proxy learning method to mitigate the modality gap between images and text. The convergence of both online strategies can be theoretically guaranteed. By combining the predicted label from the online label learning and proxy learning, our online zero-shot transfer method (OnZeta) achieves $78.94\%$ accuracy on ImageNet without accessing the entire data set. Moreover, extensive experiments on other 13 downstream tasks with different vision encoders show a more than $3\%$ improvement on average, which demonstrates the effectiveness of our proposal. Code is available at \url{https://github.com/idstcv/OnZeta}.

arXiv:2408.12248

PRG: Prompt-Based Distillation Without Annotation via Proxy Relational Graph

In this paper, we propose a new distillation method for extracting knowledge from Large Foundation Models (LFM) into lightweight models, introducing a novel supervision mode that does not require manually annotated data. While LFMs exhibit exceptional zero-shot classification abilities across datasets, relying solely on LFM-generated embeddings for distillation poses two main challenges: LFM's task-irrelevant knowledge and the high density of features. The transfer of task-irrelevant knowledge could compromise the student model's discriminative capabilities, and the high density of features within target domains obstructs the extraction of discriminative knowledge essential for the task. To address this issue, we introduce the Proxy Relational Graph (PRG) method. We initially extract task-relevant knowledge from LFMs by calculating a weighted average of logits obtained through text prompt embeddings. Then we construct sample-class proxy graphs for LFM and student models, respectively, to model the correlation between samples and class proxies. Then, we achieve the distillation of selective knowledge by aligning the relational graphs produced by both the LFM and the student model. Specifically, the distillation from LFM to the student model is achieved through two types of alignment: 1) aligning the sample nodes produced by the student model with those produced by the LFM, and 2) aligning the edge relationships in the student model's graph with those in the LFM's graph. Our experimental results validate the effectiveness of PRG, demonstrating its ability to leverage the extensive knowledge base of LFMs while skillfully circumventing their inherent limitations in focused learning scenarios. Notably, in our annotation-free framework, PRG achieves an accuracy of 76.23\% (T: 77.9\%) on CIFAR-100 and 72.44\% (T: 75.3\%) on the ImageNet-1K.