ArtEmis: Affective Language for Visual Art

Achlioptas, Panos; Ovsjanikov, Maks; Haydarov, Kilichbek; Elhoseiny, Mohamed; Guibas, Leonidas J.

doi:10.1109/cvpr46437.2021.01140

Cited by 89 publications

(81 citation statements)

References 47 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…of fine-art [32]. A related field models image aesthetics using votes on social media [23], and recently the link between style and emotion is explored [1,24]. Generative adversarial networks (GAN) such as cycle-consistent GAN [40] have been trained to map images from one domain to another, including between styles, and require labelled sets of (unpaired) images.…”

Section: Related Workmentioning

confidence: 99%

“…The consensus was determined using a graphbased vote pooling method in which edges coded by an affinity matrix A i,j reflecting the number of times both images {i, j} were simultaneously selected within an annotated cluster/moodboard. Thresholding A i,j at a given consensus level C N = [1,5] partitions the group into subgroups. Fig.…”

Section: Cleaning Style Groups (Bam-fg-c N )mentioning

confidence: 99%

“…BAM-FG-C N is crowd-annotated to form sub-groups as described in Section 3.3. The data is thresholded at different consensus levels C N = [1,5] (groupings referred to as BAM-FG-C 1..5 ). Fine-grained retrieval is evaluated over the 78K images test of BAM − F G − C 3 ; we use C 3 as the majority consensus level with the highest data volume.…”

Section: Datasets and Partitionsmentioning

confidence: 99%

“…We further explore image retrieval using the models Consensus ResNet50 [12] Ours (ALADIN) Ours (Fused) DML-BAMX [6] C N P@1 P@5 P@10 P@1 P@5 P@10 P@1 P@5 P@10 P@1 P@5 P@10 5. Image retrieval accuracy via Amazon Mechanical Turk (AMT) study; Precision (P@k) at rank k= [1,5,10] over 1000 queries on BAM-FG. (c) indicates "held out" evaluation i.e.…”

Section: User Evaluation Of Style Similaritymentioning

confidence: 99%

See 3 more Smart Citations

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

Ruta¹,

Motiian²,

Faieta³

et al. 2021

Preprint

View full text Add to dashboard Cite

Figure 1. ALADIN enables style-based visual search, by learning a fine-grained embedding for artistic style similarity. Left: Three queries (blue) and their results i.e. nearest neighbours in the ALADIN search embedding. Each result set exhibits fine-grained style coherence; a consistent style of sketching. Right (red box): a search in a prior style search embedding [6] returns results with only coarse-grained style coherence; all are different variants of watercolor style.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Cleaning Style Groups (Bam-fg-c N )mentioning

confidence: 99%

Section: Datasets and Partitionsmentioning

confidence: 99%

Section: User Evaluation Of Style Similaritymentioning

confidence: 99%

See 2 more Smart Citations

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

Ruta¹,

Motiian²,

Faieta³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…This is exacerbated when we wish to be able to specify multiple conditions, as there are even fewer training images available for each combination of conditions. We train our GAN using an enriched version of the ArtEmis dataset by Achlioptas et al [4] and investigate the effect of multiconditional labels. Two example images produced by our models can be seen in Fig.…”

Section: Introductionmentioning

confidence: 99%

Art Creation with Multi-Conditional StyleGANs

Dobler¹,

Hübscher²,

Westphal³

et al. 2022

Preprint

View full text Add to dashboard Cite

Creating meaningful art is often viewed as a uniquely human endeavor. A human artist needs a combination of unique skills, understanding, and genuine intention to create artworks that evoke deep feelings and emotions. In this paper, we introduce a multi-conditional Generative Adversarial Network (GAN) approach trained on large amounts of human paintings to synthesize realistic-looking paintings that emulate human art. Our approach is based on the StyleGAN neural network architecture, but incorporates a custom multi-conditional control mechanism that provides fine-granular control over characteristics of the generated paintings, e.g., with regard to the perceived emotion evoked in a spectator. For better control, we introduce the conditional truncation trick, which adapts the standard truncation trick for the conditional setting and diverse datasets. Finally, we develop a diverse set of evaluation techniques tailored to multi-conditional generation.

show abstract