ALADIN: All Layer Adaptive Instance Normalization for Fine-grained Style Similarity

Ruta, Dan; Motiian, Saeid; Faieta, Baldo; Lin, Zhe; Jin, Hailin; Filipkowski, Alex; Gilbert, Andrew; Collomosse, John

doi:10.1109/iccv48922.2021.01171

Cited by 25 publications

(14 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…(2) Style: Our data, described in Section 4, is derived from an artistic domain, which motivates us to consider the style view. We use the outputs of the ALADIN architecture [25], which was developed to retrieve images based on artistic style similarity. (3) Color: Since our setting is one of visual discovery, we consider color due to its important role in image retrieval.…”

Section: View-specific Representation Learningmentioning

confidence: 99%

“…And, 𝛼 𝑚 is computed as in Equation 8and 𝑠𝑖𝑚 (a, b) is a measure of similarity between a and b. For input style and color representations, we use the inverse of the 𝐿2 distance between a and b as the distance measure [18,25]. For other representations, we use 𝑠𝑖𝑚 (a, b) = a • b.…”

Section: Intent Computationmentioning

confidence: 99%

“…To enable retrieval along specialized notions of image similarity, multiple image feature extractors have been developed. Some examples include shapes within content [22], co-occurrences of objects and their relationships [17], or styles [25]. We build on existing image representation methods (e.g.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Self-supervised Multi-view Disentanglement for Expansion of Visual Collections

Jain¹,

Vaddamanu²,

Maheshwari³

et al. 2023

Preprint

View full text Add to dashboard Cite

Figure 1: Left: query collection containing a set of images. Right: each row is a ranked list of images that match the query using three notions of image similarity -objects, style, color composition -each of which we refer to as a 'view'. The top row weighs the views equally. The bottom row (our approach) weighs each view proportional to the inferred intent of the query collection. This enhances relevance (along the primary view -objects) and diversity (along other views -style and color).

show abstract

Section: View-specific Representation Learningmentioning

confidence: 99%

Section: Intent Computationmentioning

confidence: 99%

See 1 more Smart Citation

Self-supervised Multi-view Disentanglement for Expansion of Visual Collections

Jain¹,

Vaddamanu²,

Maheshwari³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…For instance, most image search and recommendation systems that are used to catalog creative content rely on algorithms to sort and organize images. In particular, these systems often use “style similarity” models to parse images (Anderson et al, 2020; Ruta et al, 2021; Wang et al, 2015). This computer vision approach detects consistencies in visual attributes between images, clustering images that it deems similar to one another.…”

Section: Background and Related Workmentioning

confidence: 99%

In the eye of the beholder: A viewer-defined conception of online visual creativity

Herman

Hwang

2022

New Media & Society

View full text Add to dashboard Cite

Despite substantial interest in developing theoretical models and technology for creativity enhancement, existing creativity research across various fields lacks a user-centered definition of creativity that can be operationalized in today’s digital spaces. To address this, we conducted a mixed-methods longitudinal research on a study website mirroring content from Bēhance, a popular online platform for creatives. Specifically, we examined how content creators and consumers explored and reflected on online creative content through textual, visual, quantitative, and behavioral data. Analyzing and triangulating these multiple data streams, we conceptualize creativity from the perspectives of its genuine “users,” the viewers. Collectively, we highlight (1) constructs of creativity that have not been emphasized in the existing literature, (2) the impact of users’ roles on content exploration and conception of creativity, and (3) the difference between machine and human users’ perception of creative content. We discuss theoretical and practical implications accordingly.

show abstract

“…Indeed, in this paper, each dataset has its own individual annotation style that results from the label generation process of each dataset. Adaptive Normalization-Based Methods For completeness, we briefly touch on the use of adaptive normalization methods, which have proven useful in previous bias adaptation and style transfer tasks (Chen et al, 2021;Nam and Kim, 2018;Karani et al, 2018;Komatsu and Gonsalves, 2022;Kim et al, 2020;Ruta et al, 2021;Jacenków et al, 2020). Several papers focus on natural imaging problems, such as artistic style transfer and image denoising (Chen et al, 2021;Nam and Kim, 2018;Komatsu and Gonsalves, 2022;Kim et al, 2020;Ruta et al, 2021).…”

Section: Background and Related Workmentioning

confidence: 99%

Rethinking Generalization: The Impact of Annotation Style on Medical Image Segmentation

Nichyporuk¹,

Cardinell²,

Szeto³

et al. 2022

Preprint

View full text Add to dashboard Cite

Generalization is an important attribute of machine learning models, particularly for those that are to be deployed in a medical context, where unreliable predictions can have real world consequences. While the failure of models to generalize across datasets is typically attributed to a mismatch in the data distributions, performance gaps are often a consequence of biases in the "ground-truth" label annotations. This is particularly important in the context of medical image segmentation of pathological structures (e.g. lesions), where the annotation process is much more subjective, and affected by a number underlying factors, including the annotation protocol, rater education/experience, and clinical aims, among others. In this paper, we show that modeling annotation biases, rather than ignoring them, poses a promising way of accounting for differences in annotation style across datasets. To this end, we propose a generalized conditioning framework to (1) learn and account for different annotation styles across multiple datasets using a single model, (2) identify similar annotation styles across different datasets in order to permit their effective aggregation, and (3) fine-tune a fully trained model to a new annotation style with just a few samples. Next, we present an image-conditioning approach to model annotation styles * . Contributed Equally ©2022 Nichyporuk et al. License: CC-BY 4.0.

show abstract

ALADIN: All Layer Adaptive Instance Normalization for Fine-grained Style Similarity

Cited by 25 publications

References 18 publications

Self-supervised Multi-view Disentanglement for Expansion of Visual Collections

Self-supervised Multi-view Disentanglement for Expansion of Visual Collections

In the eye of the beholder: A viewer-defined conception of online visual creativity

Rethinking Generalization: The Impact of Annotation Style on Medical Image Segmentation

Contact Info

Product

Resources

About