no code implementations • 15 Apr 2024 • Masahito Toba, Seiichi Uchida, Hideaki Hayashi
In pseudo-labeling (PL), which is a type of semi-supervised learning, pseudo-labels are assigned based on the confidence scores provided by the classifier; therefore, accurate confidence is important for successful PL.
no code implementations • 19 Mar 2024 • Daichi Haraguchi, Wataru Shimoda, Kota Yamaguchi, Seiichi Uchida
Second, it is demonstrated that the disentangled features produced by total disentanglement apply to a variety of tasks, including font recognition, character recognition, and one-shot font image generation.
1 code implementation • 6 Mar 2024 • Takahiro Shirakawa, Seiichi Uchida
The current layout-aware text-to-image diffusion models still have several issues, including mismatches between the text and layout conditions and quality degradation of generated images.
no code implementations • 5 Mar 2024 • Sho Shimotsumagari, Shumpei Takezaki, Daichi Haraguchi, Seiichi Uchida
By applying machine-printed and handwritten character images to the two modalities, CycleDM realizes the conversion between them.
no code implementations • 1 Mar 2024 • Shumpei Takezaki, Seiichi Uchida
Diffusion models have recently been used for medical image generation because of their high image quality.
no code implementations • 26 Feb 2024 • Yugo Kubota, Daichi Haraguchi, Seiichi Uchida
However, the correlation between fonts and their impression is weak and unstable because impressions are subjective.
no code implementations • 26 Feb 2024 • Daichi Haraguchi, Brian Kenji Iwana, Seiichi Uchida
In the experiment, we found that semantic information is sufficient to determine the genre; however, text design is helpful in adding more discriminative features for book genres.
no code implementations • 23 Feb 2024 • Kazuki Kitajima, Daichi Haraguchi, Seiichi Uchida
To realize stable impression estimation even with such a dataset, we propose an exemplar-based impression estimation approach, which relies on a strategy of ensembling impressions of exemplar fonts that are similar to the input image.
no code implementations • 22 Feb 2024 • Tetta Kondo, Shumpei Takezaki, Daichi Haraguchi, Seiichi Uchida
In this paper, we employ diffusion models to generate new font styles by interpolating a pair of reference fonts with different styles.
no code implementations • 22 Feb 2024 • Kei Nakatsuru, Seiichi Uchida
Kerning is the task of setting appropriate horizontal spaces for all possible letter pairs of a certain font.
no code implementations • 22 Feb 2024 • KhayTze Peong, Seiichi Uchida, Daichi Haraguchi
The proposed system is a novel combination of two off-the-shelf methods for diffusion models, ControlNet and Blended Latent Diffusion.
no code implementations • 20 Oct 2023 • Yuya Saito, Shinnosuke Matsuo, Seiichi Uchida, Daiki Suehiro
This paper tackles the problem of the worst-class error rate, instead of the standard error rate averaged over all classes.
no code implementations • 10 Oct 2023 • Daichi Haraguchi, Seiichi Uchida
This paper proposes an attention mechanism to find important local parts.
1 code implementation • 13 Sep 2023 • Shinnosuke Matsuo, Xiaomeng Wu, Gantugs Atarsaikhan, Akisato Kimura, Kunio Kashino, Brian Kenji Iwana, Seiichi Uchida
Unlike other learnable models using DTW for warping, our model predicts all local correspondences between two time series and is trained based on metric learning, which enables it to learn the optimal data-dependent warping for the target task.
no code implementations • 5 Sep 2023 • Wataru Shimoda, Daichi Haraguchi, Seiichi Uchida, Kota Yamaguchi
In this work, we consider the typography generation task that aims at producing diverse typographic styling for the given graphic document.
1 code implementation • 1 Sep 2023 • Hayato Mitani, Akisato Kimura, Seiichi Uchida
Scene text removal (STR) is the image transformation task to remove text regions in scene images.
1 code implementation • 21 Jun 2023 • Takahiro Shirakawa, Seiichi Uchida
For example, the pair of 'A' and 'V' shows a high ambigramability (that is, it is easy to generate their ambigrams), and the pair of 'D' and 'K' shows a lower ambigramability.
no code implementations • 21 Jun 2023 • Naoya Yasukochi, Hideaki Hayashi, Daichi Haraguchi, Seiichi Uchida
There are various font styles in the world.
1 code implementation • 16 Jun 2023 • Guangtao Lyu, Kun Liu, Anna Zhu, Seiichi Uchida, Brian Kenji Iwana
To tackle these problems, we propose a novel Feature Erasing and Transferring (FET) mechanism to reconfigure the encoded features for STR in this paper.
no code implementations • 27 Apr 2023 • Yusuke Nagata, Brian Kenji Iwana, Seiichi Uchida
We propose a Transformer-based method to solve this problem and show the results of the typeface contour completion.
1 code implementation • 12 Mar 2023 • Prakash Chandra Chhipa, Muskaan Chopra, Gopal Mengi, Varun Gupta, Richa Upadhyay, Meenakshi Subhash Chippa, Kanjar De, Rajkumar Saini, Seiichi Uchida, Marcus Liwicki
This work investigates the unexplored usability of self-supervised representation learning in the direction of functional knowledge transfer.
no code implementations • 2 Mar 2023 • Shota Harada, Ryoma Bise, Kengo Araki, Akihiko Yoshizawa, Kazuhiro Terada, Mariyo Kurata, Naoki Nakajima, Hiroyuki Abe, Tetsuo Ushiku, Seiichi Uchida
Semi-supervised domain adaptation is a technique to build a classifier for a target domain by modifying a classifier in another (source) domain using many unlabeled samples and a small number of labeled samples from the target domain.
no code implementations • 24 Feb 2023 • Shumpei Takezaki, Kiyohito Tanaka, Seiichi Uchida, Takeaki Kadota
We propose continuous DA as a solution to the two issues.
1 code implementation • 17 Feb 2023 • Shinnosuke Matsuo, Ryoma Bise, Seiichi Uchida, Daiki Suehiro
This paper proposes a novel and efficient method for Learning from Label Proportions (LLP), whose goal is to train a classifier only by using the class label proportions of instance sets, called bags.
1 code implementation • 18 Oct 2022 • Prakash Chandra Chhipa, Richa Upadhyay, Rajkumar Saini, Lars Lindqvist, Richard Nordenskjold, Seiichi Uchida, Marcus Liwicki
This work presents a novel self-supervised representation learning method to learn efficient representations without labels on images from a 3DPM sensor (3-Dimensional Particle Measurement; estimates the particle size distribution of material) utilizing RGB images and depth maps of mining material on the conveyor belt.
no code implementations • 5 Aug 2022 • Takeaki Kadota, Hideaki Hayashi, Ryoma Bise, Kiyohito Tanaka, Seiichi Uchida
This paper proposes a deep Bayesian active-learning-to-rank, which trains a Bayesian convolutional neural network while automatically selecting appropriate pairs for relative annotation.
1 code implementation • 31 May 2022 • Chean Fei Shee, Seiichi Uchida
A multi-layer image is more valuable than a single-layer image from a graphic designer's perspective.
no code implementations • 19 Mar 2022 • Seiya Matsuda, Akisato Kimura, Seiichi Uchida
Our goal is to generate fonts with specific impressions, by training a generative adversarial network with a font dataset with impression labels.
no code implementations • 17 Mar 2022 • Xiaotong Ji, Yuchen Zheng, Daiki Suehiro, Seiichi Uchida
The highlights of LwR are: (1) the rejection strategy is not heuristic but has a strong background from a machine learning theory, and (2) the rejection function can be trained on an arbitrary feature space which is different from the feature space for classification.
no code implementations • 17 Mar 2022 • Xiaotong Ji, Yan Zheng, Daiki Suehiro, Seiichi Uchida
Signature verification, as a crucial practical documentation analysis task, has been continuously studied by researchers in machine learning and pattern recognition fields.
1 code implementation • 15 Mar 2022 • Prakash Chandra Chhipa, Richa Upadhyay, Gustav Grund Pihlgren, Rajkumar Saini, Seiichi Uchida, Marcus Liwicki
This work presents a novel self-supervised pre-training method to learn efficient representations without labels on histopathology medical images utilizing magnification factors.
Ranked #1 on Breast Cancer Histology Image Classification on BreakHis (Accuracy (Inter-Patient) metric)
Breast Cancer Histology Image Classification (20% labels) Classification Of Breast Cancer Histology Images +2
no code implementations • 11 Mar 2022 • Masaya Ueda, Akisato Kimura, Seiichi Uchida
The versatility of Transformer allows us to realize two very different approaches for the analysis, i. e., multi-label classification and translation.
1 code implementation • 10 Mar 2022 • Yusuke Nagata, Jinki Otao, Daichi Haraguchi, Seiichi Uchida
The outline format, such as TrueType, represents each character as a sequence of control points of stroke contours and is frequently used in born-digital documents.
no code implementations • 6 Nov 2021 • Shota Harada, Ryoma Bise, Hideaki Hayashi, Kiyohito Tanaka, Seiichi Uchida
Ulcerative colitis (UC) classification, which is an important task for endoscopic diagnosis, involves two main difficulties.
1 code implementation • ICCV 2021 • Wataru Shimoda, Daichi Haraguchi, Seiichi Uchida, Kota Yamaguchi
Editing raster text is a promising but challenging task.
no code implementations • 29 Jun 2021 • Kaigen Tsuji, Seiichi Uchida, Brian Kenji Iwana
In this paper, we attempt to specifically find the trends in font usage using robust regression on a large collection of text images.
1 code implementation • 24 May 2021 • Wensheng Zhang, Yan Zheng, Taiga Miyazono, Seiichi Uchida, Brian Kenji Iwana
Book covers are intentionally designed and provide an introduction to a book.
1 code implementation • 19 May 2021 • Taiga Miyazono, Brian Kenji Iwana, Daichi Haraguchi, Seiichi Uchida
We propose an end-to-end neural network that inputs the book cover, a target location mask, and a desired book title and outputs stylized text suitable for the cover.
Ranked #1 on Font Generation on Book Cover Dataset
no code implementations • 1 Apr 2021 • Shintaro Nishi, Takeaki Kadota, Seiichi Uchida
Various findings include the weak positive correlation between the text area ratio and the number of followers of the company.
1 code implementation • 28 Mar 2021 • Shinnosuke Matsuo, Xiaomeng Wu, Gantugs Atarsaikhan, Akisato Kimura, Kunio Kashino, Brian Kenji Iwana, Seiichi Uchida
This approach adapts a parameterized attention model to time warping for greater and more adaptive temporal invariance.
no code implementations • 26 Mar 2021 • Masaya Ueda, Akisato Kimura, Seiichi Uchida
Various fonts give different impressions, such as legible, rough, and comic-text. This paper aims to analyze the correlation between the local shapes, or parts, and the impression of fonts.
no code implementations • 23 Mar 2021 • Jihun Kang, Daichi Haraguchi, Seiya Matsuda, Akisato Kimura, Seiichi Uchida
The difficulty is that the impression words attached to a font are often very noisy.
no code implementations • 18 Mar 2021 • Seiya Matsuda, Akisato Kimura, Seiichi Uchida
Various fonts give us various impressions, which are often represented by words.
1 code implementation • 17 Mar 2021 • Takato Otsuzuki, Heon Song, Seiichi Uchida, Hideaki Hayashi
As part of our framework, a parameterized pooling layer is proposed in which the kernel shape and pooling operation are trainable using two parameters, thereby allowing flexible pooling of the input data.
no code implementations • 8 Mar 2021 • Shinnosuke Matsuo, Seiichi Uchida, Brian Kenji Iwana
To exploit this fact, we propose the use of self-augmentation and combine it with multi-modal feature embedding.
no code implementations • 26 Feb 2021 • Shohei Kubota, Hideaki Hayashi, Tomohiro Hayase, Seiichi Uchida
The interpretability of neural networks (NNs) is a challenging but essential topic for transparency in the decision-making process using machine learning.
no code implementations • 27 Sep 2020 • Masaki Yamagata, Hideaki Hayashi, Seiichi Uchida
In this paper, we propose a temporal prediction model that can deal with this bifurcation structure.
no code implementations • 23 Sep 2020 • Keisuke Kanda, Brian Kenji Iwana, Seiichi Uchida
In this study, we use a reinforcement learning (RL) framework to realize handwriting generation with the careful future planning ability.
2 code implementations • 19 Sep 2020 • Heon Song, Daiki Suehiro, Seiichi Uchida
For visual object tracking, it is difficult to realize an almighty online tracker due to the huge variations of target appearance depending on an image sequence.
Ranked #1 on Visual Object Tracking on TempleColor128
1 code implementation • 31 Jul 2020 • Brian Kenji Iwana, Seiichi Uchida
In this paper, we survey data augmentation techniques for time series and their application to time series classification with neural networks.
no code implementations • 30 Jul 2020 • Koki Takeshita, Juntaro Shioyama, Seiichi Uchida
Our daily life is surrounded by textual information.
no code implementations • 22 Jun 2020 • Kohei Baba, Seiichi Uchida, Brian Kenji Iwana
The purpose of this paper is to reveal the ability that Convolutional Neural Networks (CNN) have on the novel task of image-to-image language conversion.
no code implementations • 21 Jun 2020 • Shota Sakaguchi, Jun Kato, Masataka Goto, Seiichi Uchida
In order to analyze the motion of lyric words, we first apply a state-of-the-art scene text detector and recognizer to each video frame.
1 code implementation • PLOS ONE 2020 • Gantugs Atarsaikhan, Brian Kenji Iwana, Seiichi Uchida
Designing logos, typefaces, and other decorated shapes can require professional skills.
no code implementations • 6 May 2020 • Takato Otsuzuki, Hideaki Hayashi, Yuchen Zheng, Seiichi Uchida
This means that max pooling is too flexible to compensate for actual deformations.
2 code implementations • 19 Apr 2020 • Brian Kenji Iwana, Seiichi Uchida
In order to address this problem, we propose a novel time series data augmentation called guided warping.
no code implementations • 18 Apr 2020 • Masaya Ikoma, Brian Kenji Iwana, Seiichi Uchida
In natural scenes and documents, we can find the correlation between a text and its color.
no code implementations • 7 Apr 2020 • Takuro Karamatsu, Gibran Benitez-Garcia, Keiji Yanai, Seiichi Uchida
In this paper, we tackle a challenging domain conversion task between photo and icon images.
1 code implementation • 24 Jan 2020 • Daichi Haraguchi, Shota Harada, Brian Kenji Iwana, Yuto Shinahara, Seiichi Uchida
Moreover, we analyzed the relationship between character classes and font identification accuracy.
no code implementations • 21 Jan 2020 • Gantugs Atarsaikhan, Brian Kenji Iwana, Seiichi Uchida
In our proposed method, the difference of font styles between two different fonts is found and transferred to another font using neural style transfer.
no code implementations • ICLR 2021 • Hideaki Hayashi, Seiichi Uchida
We propose a sparse classifier based on a discriminative GMM, referred to as a sparse discriminative Gaussian mixture (SDGM).
no code implementations • 25 Sep 2019 • Hideaki Hayashi, Seiichi Uchida
In the SDGM, a GMM-based discriminative model is trained by sparse Bayesian learning.
1 code implementation • 6 Aug 2019 • Brian Kenji Iwana, Ryohei Kuroki, Seiichi Uchida
Through qualitative and quantitative analysis, we demonstrate that SGLRP can successfully localize and attribute the regions on input images which contribute to a target object's classification.
no code implementations • 24 Jun 2019 • Yuto Shinahara, Takuro Karamatsu, Daisuke Harada, Kota Yamaguchi, Seiichi Uchida
In this paper, we conduct a large-scale study of font statistics in book covers and online advertisements.
no code implementations • 17 Jun 2019 • Toshiki Nakamura, Anna Zhu, Seiichi Uchida
In this paper, we design the scene text magnifier through interacted four CNN-based networks: character erasing, character extraction, character magnify, and image synthesis.
no code implementations • 14 Jun 2019 • Taichi Sumi, Brian Kenji Iwana, Hideaki Hayashi, Seiichi Uchida
This research attempts to construct a network that can convert online and offline handwritten characters to each other.
no code implementations • 30 May 2019 • Hideaki Hayashi, Seiichi Uchida
In this paper, we propose a trainable multiplication layer (TML) for a neural network that can be used to calculate the multiplication between the input features.
1 code implementation • 29 May 2019 • Hideaki Hayashi, Kohtaro Abe, Seiichi Uchida
In GlyphGAN, the input vector for the generator network consists of two vectors: character class vector and style vector.
1 code implementation • 26 May 2019 • Kumar Shridhar, Joonho Lee, Hideaki Hayashi, Purvanshi Mehta, Brian Kenji Iwana, Seokjun Kang, Seiichi Uchida, Sheraz Ahmed, Andreas Dengel
We show that ProbAct increases the classification accuracy by +2-3% compared to ReLU or other conventional activation functions on both original datasets and when datasets are reduced to 50% and 25% of the original size.
no code implementations • 17 May 2019 • Shota Harada, Hideaki Hayashi, Seiichi Uchida
GAN-based generative models only learn the projection between a random distribution as input data and the distribution of training data. Therefore, the relationship between input and generated data is unclear, and the characteristics of the data generated from this model cannot be controlled.
no code implementations • 25 Aug 2018 • Shailza Jolly, Brian Kenji Iwana, Ryohei Kuroki, Seiichi Uchida
We use LRP to explain the pixel-wise contributions of book cover design and highlight the design elements contributing towards particular genres.
1 code implementation • 2 Mar 2018 • Gantugs Atarsaikhan, Brian Kenji Iwana, Seiichi Uchida
We propose using neural style transfer with clip art and text for the creation of new and genuine logos.
no code implementations • 18 Dec 2017 • Brian Kenji Iwana, Seiichi Uchida
In this paper, we propose a method of improving temporal Convolutional Neural Networks (CNN) by determining the optimal alignment of weights and inputs using dynamic programming.
no code implementations • 6 Dec 2017 • Frédéric Rayar, Masanori Goto, Seiichi Uchida
In this paper, we present a study on sample preselection in large training data set for CNN-based classification.
no code implementations • 8 May 2017 • Toshiki Nakamura, Anna Zhu, Keiji Yanai, Seiichi Uchida
That proves the effectiveness of proposed method for erasing the text in natural scene images.
no code implementations • 20 Jan 2017 • Tomo Miyazaki, Tatsunori Tsuchiya, Yoshihiro Sugaya, Shinichiro Omachi, Masakazu Iwamura, Seiichi Uchida, Koichi Kise
The proposed method uses strokes from given samples for font generation.
no code implementations • 25 Dec 2016 • Jinho Lee, Brian Kenji Iwana, Shouta Ide, Seiichi Uchida
Thus, we propose a new and robust tracking method using a Fully Convolutional Network (FCN) to obtain an object probability map and Dynamic Programming (DP) to seek the globally optimal path through all frames of video.
4 code implementations • 28 Oct 2016 • Brian Kenji Iwana, Syed Tahseen Raza Rizvi, Sheraz Ahmed, Andreas Dengel, Seiichi Uchida
Book covers communicate information to potential readers, but can that same information be learned by computers?
Ranked #1 on Genre classification on Book Cover Dataset