Kuniaki Saito

Kuniaki Saito, Kihyuk Sohn, Xiang Zhang et al.

In Composed Image Retrieval (CIR), a user combines a query image with text to describe their intended target. Existing methods rely on supervised learning of CIR models using labeled triplets consisting of the query image, text specification, and the target image. Labeling such triplets is expensive and hinders broad applicability of CIR. In this work, we propose to study an important task, Zero-Shot Composed Image Retrieval (ZS-CIR), whose goal is to build a CIR model without requiring labeled triplets for training. To this end, we propose a novel method, called Pic2Word, that requires only weakly labeled image-caption pairs and unlabeled image datasets to train. Unlike existing supervised CIR models, our model trained on weakly labeled or unlabeled datasets shows strong generalization across diverse ZS-CIR tasks, e.g., attribute editing, object composition, and domain conversion. Our approach outperforms several supervised CIR methods on the common CIR benchmark, CIRR and Fashion-IQ. Code will be made publicly available at https://github.com/google-research/composed_image_retrieval.

Piotr Teterwak, Kuniaki Saito, Theodoros Tsiligkaridis et al.

Domain Generalization (DG) aims to develop classifiers that can generalize to new, unseen data distributions, a critical capability when collecting new domain-specific data is impractical. A common DG baseline minimizes the empirical risk on the source domains. Recent studies have shown that this approach, known as Empirical Risk Minimization (ERM), can outperform most more complex DG methods when properly tuned. However, these studies have primarily focused on a narrow set of hyperparameters, neglecting other factors that can enhance robustness and prevent overfitting and catastrophic forgetting, properties which are critical for strong DG performance. In our investigation of training data utilization (i.e., duration and setting validation splits), initialization, and additional regularizers, we find that tuning these previously overlooked factors significantly improves model generalization across diverse datasets without adding much complexity. We call this improved, yet simple baseline ERM++. Despite its ease of implementation, ERM++ improves DG performance by over 5\% compared to prior ERM baselines on a standard benchmark of 5 datasets with a ResNet-50 and over 15\% with a ViT-B/16. It also outperforms all state-of-the-art methods on DomainBed datasets with both architectures. Importantly, ERM++ is easy to integrate into existing frameworks like DomainBed, making it a practical and powerful tool for researchers and practitioners. Overall, ERM++ challenges the need for more complex DG methods by providing a stronger, more reliable baseline that maintains simplicity and ease of use. Code is available at \url{https://github.com/piotr-teterwak/erm_plusplus}

Kuniaki Saito, Donghyun Kim, Piotr Teterwak et al.

Building object detectors that are robust to domain shifts is critical for real-world applications. Prior approaches fine-tune a pre-trained backbone and risk overfitting it to in-distribution (ID) data and distorting features useful for out-of-distribution (OOD) generalization. We propose to use Relative Gradient Norm (RGN) as a way to measure the vulnerability of a backbone to feature distortion, and show that high RGN is indeed correlated with lower OOD performance. Our analysis of RGN yields interesting findings: some backbones lose OOD robustness during fine-tuning, but others gain robustness because their architecture prevents the parameters from changing too much from the initial model. Given these findings, we present recipes to boost OOD robustness for both types of backbones. Specifically, we investigate regularization and architectural choices for minimizing gradient updates so as to prevent the tuned backbone from losing generalizable features. Our proposed techniques complement each other and show substantial improvements over baselines on diverse architectures and datasets. Code is available at https://github.com/VisionLearningGroup/mind_back.

Kuniaki Saito, Kihyuk Sohn, Xiang Zhang et al.

Image-classification datasets have been used to pretrain image recognition models. Recently, web-scale image-caption datasets have emerged as a source of powerful pretraining alternative. Image-caption datasets are more ``open-domain'', containing a wider variety of scene types and vocabulary words than traditional classification datasets, and models trained on these datasets have demonstrated strong performance on few- and zero-shot recognition tasks. When naively unifying image-classification and -caption dataset, we show that such dataset biases negatively affect pre-training by reducing the generalizability of learned representations and thus jeopardizing zero-shot performance since the unification can tailor the model for the classification dataset, making it vulnerable to the distribution shift from the dataset. In this work, we address the problem by disentangling the dataset bias using prefix tokens that inform a language encoder of the type of the input dataset (e.g., image-classification or caption) at training time. This approach allows the language encoder to share the knowledge from two datasets as well as switch the mode of feature extraction, i.e., image-classification dataset or image-caption dataset tailored mode, where we use image-caption mode in the zero-shot evaluation. Our method is generic and can be easily integrated into existing VL pre-training objectives such as CLIP or UniCL. In experiments, we show that this simple technique improves the performance in zero-shot image recognition accuracy and robustness to the image-level distribution shift.

Kwanyong Park, Kuniaki Saito, Donghyun Kim

Vision-language (VL) models often exhibit a limited understanding of complex expressions of visual objects (e.g., attributes, shapes, and their relations), given complex and diverse language queries. Traditional approaches attempt to improve VL models using hard negative synthetic text, but their effectiveness is limited. In this paper, we harness the exceptional compositional understanding capabilities of generative foundational models. We introduce a novel method for structured synthetic data generation aimed at enhancing the compositional understanding of VL models in language-based object detection. Our framework generates densely paired positive and negative triplets (image, text descriptions, and bounding boxes) in both image and text domains. By leveraging these synthetic triplets, we transform 'weaker' VL models into 'stronger' models in terms of compositional understanding, a process we call "Weak-to-Strong Compositional Learning" (WSCL). To achieve this, we propose a new compositional contrastive learning formulation that discovers semantics and structures in complex descriptions from synthetic triplets. As a result, VL models trained with our synthetic data generation exhibit a significant performance boost in the Omnilabel benchmark by up to +5AP and the D3 benchmark by +6.9AP upon existing baselines.

Risa Shinoda, Kaede Shiohara, Nakamasa Inoue et al.

Understanding animal species from multimodal data poses an emerging challenge at the intersection of computer vision and ecology. While recent biological models, such as BioCLIP, have demonstrated strong alignment between images and textual taxonomic information for species identification, the integration of the audio modality remains an open problem. We propose BioVITA, a novel visual-textual-acoustic alignment framework for biological applications. BioVITA involves (i) a training dataset, (ii) a representation model, and (iii) a retrieval benchmark. First, we construct a large-scale training dataset comprising 1.3 million audio clips and 2.3 million images, covering 14,133 species annotated with 34 ecological trait labels. Second, building upon BioCLIP2, we introduce a two-stage training framework to effectively align audio representations with visual and textual representations. Third, we develop a cross-modal retrieval benchmark that covers all possible directional retrieval across the three modalities (i.e., image-to-audio, audio-to-text, text-to-image, and their reverse directions), with three taxonomic levels: Family, Genus, and Species. Extensive experiments demonstrate that our model learns a unified representation space that captures species-level semantics beyond taxonomy, advancing multimodal biodiversity understanding. The project page is available at: https://dahlian00.github.io/BioVITA_Page/

Kuniaki Saito, Risa Shinoda, Shohei Tanaka et al.

Hallucination detection in captions (HalDec) assesses a vision-language model's ability to correctly align image content with text by identifying errors in captions that misrepresent the image. Beyond evaluation, effective hallucination detection is also essential for curating high-quality image-caption pairs used to train VLMs. However, the generalizability of VLMs as hallucination detectors across different captioning models and hallucination types remains unclear due to the lack of a comprehensive benchmark. In this work, we introduce HalDec-Bench, a benchmark designed to evaluate hallucination detectors in a principled and interpretable manner. HalDec-Bench contains captions generated by diverse VLMs together with human annotations indicating the presence of hallucinations, detailed hallucination-type categories, and segment-level labels. The benchmark provides tasks with a wide range of difficulty levels and reveals performance differences across models that are not visible in existing multimodal reasoning or alignment benchmarks. Our analysis further uncovers two key findings. First, detectors tend to recognize sentences appearing at the beginning of a response as correct, regardless of their actual correctness. Second, our experiments suggest that dataset noise can be substantially reduced by using strong VLMs as filters while employing recent VLMs as caption generators. Our project page is available at https://dahlian00.github.io/HalDec-Bench-Page/.

Hikaru Asano, Tadashi Kozuno, Kuniaki Saito et al.

Masked Diffusion Language Models (MDLMs) generate text by iteratively filling masked tokens, requiring two coupled decisions at each step: which positions to unmask (where-to-unmask) and which tokens to place (what-to-unmask). While standard MDLM training directly optimizes token prediction (what-to-unmask), inference-time unmasking orders (where-to-unmask) are typically determined by heuristic confidence measures or trained through reinforcement learning with costly on-policy rollouts. To address this, we introduce Gt-Margin, a position-wise score derived from ground-truth tokens, defined as the probability margin between the correct token and its strongest alternative. Gt-Margin yields an oracle unmasking order that prioritizes easier positions first under each partially masked state. We demonstrate that leveraging this oracle unmasking order significantly enhances final generation quality, particularly on logical reasoning benchmarks. Building on this insight, we train a supervised unmasking planner via learning-to-rank to imitate the oracle ordering from masked contexts. The resulting planner integrates into standard MDLM sampling to select where-to-unmask, improving reasoning accuracy without modifying the token prediction model.

Yi Gu, Kuniaki Saito, Jiaxin Ma

As medical diagnoses increasingly leverage multimodal data, machine learning models are expected to effectively fuse heterogeneous information while remaining robust to missing modalities. In this work, we propose a novel multimodal learning framework that integrates enhanced modalities dropout and contrastive learning to address real-world limitations such as modality imbalance and missingness. Our approach introduces learnable modality tokens for improving missingness-aware fusion of modalities and augments conventional unimodal contrastive objectives with fused multimodal representations. We validate our framework on large-scale clinical datasets for disease detection and prediction tasks, encompassing both visual and tabular modalities. Experimental results demonstrate that our method achieves state-of-the-art performance, particularly in challenging and practical scenarios where only a single modality is available. Furthermore, we show its adaptability through successful integration with a recent CT foundation model. Our findings highlight the effectiveness, efficiency, and generalizability of our approach for multimodal learning, offering a scalable, low-cost solution with significant potential for real-world clinical applications. The code is available at https://github.com/omron-sinicx/medical-modality-dropout.

Kuniaki Saito, Donghyun Kim, Kwanyong Park et al.

An image captioning model flexibly switching its language pattern, e.g., descriptiveness and length, should be useful since it can be applied to diverse applications. However, despite the dramatic improvement in generative vision-language models, fine-grained control over the properties of generated captions is not easy due to two reasons: (i) existing models are not given the properties as a condition during training and (ii) existing models cannot smoothly transition its language pattern from one state to the other. Given this challenge, we propose a new approach, CaptionSmiths, to acquire a single captioning model that can handle diverse language patterns. First, our approach quantifies three properties of each caption, length, descriptiveness, and uniqueness of a word, as continuous scalar values, without human annotation. Given the values, we represent the conditioning via interpolation between two endpoint vectors corresponding to the extreme states, e.g., one for a very short caption and one for a very long caption. Empirical results demonstrate that the resulting model can smoothly change the properties of the output captions and show higher lexical alignment than baselines. For instance, CaptionSmiths reduces the error in controlling caption length by 506\% despite better lexical alignment. Code will be available on https://github.com/omron-sinicx/captionsmiths.

Kuniaki Saito, Donghyun Kim, Piotr Teterwak et al.

Unsupervised domain adaptation (UDA) methods can dramatically improve generalization on unlabeled target domains. However, optimal hyper-parameter selection is critical to achieving high accuracy and avoiding negative transfer. Supervised hyper-parameter validation is not possible without labeled target data, which raises the question: How can we validate unsupervised adaptation techniques in a realistic way? We first empirically analyze existing criteria and demonstrate that they are not very effective for tuning hyper-parameters. Intuitively, a well-trained source classifier should embed target samples of the same class nearby, forming dense neighborhoods in feature space. Based on this assumption, we propose a novel unsupervised validation criterion that measures the density of soft neighborhoods by computing the entropy of the similarity distribution between points. Our criterion is simpler than competing validation methods, yet more effective; it can tune hyper-parameters and the number of training iterations in both image classification and semantic segmentation models. The code used for the paper will be available at \url{https://github.com/VisionLearningGroup/SND}.

Kuniaki Saito, Donghyun Kim, Stan Sclaroff et al.

Unsupervised domain adaptation methods traditionally assume that all source categories are present in the target domain. In practice, little may be known about the category overlap between the two domains. While some methods address target settings with either partial or open-set categories, they assume that the particular setting is known a priori. We propose a more universally applicable domain adaptation framework that can handle arbitrary category shift, called Domain Adaptative Neighborhood Clustering via Entropy optimization (DANCE). DANCE combines two novel ideas: First, as we cannot fully rely on source categories to learn features discriminative for the target, we propose a novel neighborhood clustering technique to learn the structure of the target domain in a self-supervised way. Second, we use entropy-based feature alignment and rejection to align target features with the source, or reject them as unknown categories based on their entropy. We show through extensive experiments that DANCE outperforms baselines across open-set, open-partial and partial domain adaptation settings. Implementation is available at https://github.com/VisionLearningGroup/DANCE.

Kuniaki Saito, Yoshitaka Ushiku, Tatsuya Harada et al.

We propose an approach for unsupervised adaptation of object detectors from label-rich to label-poor domains which can significantly reduce annotation costs associated with detection. Recently, approaches that align distributions of source and target images using an adversarial loss have been proven effective for adapting object classifiers. However, for object detection, fully matching the entire distributions of source and target images to each other at the global image level may fail, as domains could have distinct scene layouts and different combinations of objects. On the other hand, strong matching of local features such as texture and color makes sense, as it does not change category level semantics. This motivates us to propose a novel method for detector adaptation based on strong local alignment and weak global alignment. Our key contribution is the weak alignment model, which focuses the adversarial alignment loss on images that are globally similar and puts less emphasis on aligning images that are globally dissimilar. Additionally, we design the strong domain alignment model to only look at local receptive fields of the feature map. We empirically verify the effectiveness of our method on four datasets comprising both large and small domain shifts. Our code is available at \url{https://github.com/VisionLearningGroup/DA_Detection}

Kuniaki Saito, Kohei Watanabe, Yoshitaka Ushiku et al.

In this work, we present a method for unsupervised domain adaptation. Many adversarial learning methods train domain classifier networks to distinguish the features as either a source or target and train a feature generator network to mimic the discriminator. Two problems exist with these methods. First, the domain classifier only tries to distinguish the features as a source or target and thus does not consider task-specific decision boundaries between classes. Therefore, a trained generator can generate ambiguous features near class boundaries. Second, these methods aim to completely match the feature distributions between different domains, which is difficult because of each domain's characteristics. To solve these problems, we introduce a new approach that attempts to align distributions of source and target by utilizing the task-specific decision boundaries. We propose to maximize the discrepancy between two classifiers' outputs to detect target samples that are far from the support of the source. A feature generator learns to generate target features near the support to minimize the discrepancy. Our method outperforms other methods on several datasets of image classification and semantic segmentation. The codes are available at \url{https://github.com/mil-tokyo/MCD_DA}

Takara Taniguchi, Kuniaki Saito, Atsushi Hashimoto

Vision Language Models (VLMs) are increasingly deployed in autonomous vehicles and mobile systems, making it crucial to evaluate their ability to support safer decision-making in complex environments. However, existing benchmarks inadequately cover diverse hazardous situations, especially anomalous scenarios with spatio-temporal dynamics. While image editing models are a promising means to synthesize such hazards, it remains challenging to generate well-formulated scenarios that include moving, intrusive, and distant objects frequently observed in the real world. To address this gap, we introduce \textbf{HazardForge}, a scalable pipeline that leverages image editing models to generate these scenarios with layout decision algorithms, and validation modules. Using HazardForge, we construct \textbf{MovSafeBench}, a multiple-choice question (MCQ) benchmark comprising 7,254 images and corresponding QA pairs across 13 object categories, covering both normal and anomalous objects. Experiments using MovSafeBench show that VLM performance degrades notably under conditions including anomalous objects, with the largest drop in scenarios requiring nuanced motion understanding.

Kuniaki Saito, Kihyuk Sohn, Chen-Yu Lee et al.

Large language models require updates to remain up-to-date or adapt to new domains by fine-tuning them with new documents. One key is memorizing the latest information in a way that the memorized information is extractable with a query prompt. However, LLMs suffer from a phenomenon called perplexity curse; despite minimizing document perplexity during fine-tuning, LLMs struggle to extract information through a prompt sentence. In this new knowledge acquisition and extraction, we find a very intriguing fact that LLMs can accurately answer questions about the first sentence, but they struggle to extract information described in the middle or end of the documents used for fine-tuning. Our study suggests that the auto-regressive training causes this issue; each token is prompted by reliance on all previous tokens, which hinders the model from recalling information from training documents by question prompts. To conduct the in-depth study, we publish both synthetic and real datasets, enabling the evaluation of the QA performance w.r.t. the position of the corresponding answer in a document. Our investigation shows that even a large model suffers from the perplexity curse, but regularization such as denoising auto-regressive loss can enhance the information extraction from diverse positions. These findings will be (i) a key to improving knowledge extraction from LLMs and (ii) new elements to discuss the trade-off between RAG and fine-tuning in adapting LLMs to a new domain.

Piotr Teterwak, Kuniaki Saito, Theodoros Tsiligkaridis et al.

Multi-Source Domain Generalization (DG) is the task of training on multiple source domains and achieving high classification performance on unseen target domains. Recent methods combine robust features from web-scale pretrained backbones with new features learned from source data, and this has dramatically improved benchmark results. However, it remains unclear if DG finetuning methods are becoming better over time, or if improved benchmark performance is simply an artifact of stronger pre-training. Prior studies have shown that perceptual similarity to pre-training data correlates with zero-shot performance, but we find the effect limited in the DG setting. Instead, we posit that having perceptually similar data in pretraining is not enough; and that it is how well these data were learned that determines performance. This leads us to introduce the Alignment Hypothesis, which states that the final DG performance will be high if and only if alignment of image and class label text embeddings is high. Our experiments confirm the Alignment Hypothesis is true, and we use it as an analysis tool of existing DG methods evaluated on DomainBed datasets by splitting evaluation data into In-pretraining (IP) and Out-of-pretraining (OOP). We show that all evaluated DG methods struggle on DomainBed-OOP, while recent methods excel on DomainBed-IP. Put together, our findings highlight the need for DG methods which can generalize beyond pretraining alignment.

Risa Shinoda, Kuniaki Saito, Shohei Tanaka et al.

Building a large-scale figure QA dataset requires a considerable amount of work, from gathering and selecting figures to extracting attributes like text, numbers, and colors, and generating QAs. Although recent developments in LLMs have led to efforts to synthesize figures, most of these focus primarily on QA generation. Additionally, creating figures directly using LLMs often encounters issues such as code errors, similar-looking figures, and repetitive content in figures. To address this issue, we present SBSFigures (Stage-by-Stage Synthetic Figures), a dataset for pre-training figure QA. Our proposed pipeline enables the creation of chart figures with complete annotations of the visualized data and dense QA annotations without any manual annotation process. Our stage-by-stage pipeline makes it possible to create diverse topic and appearance figures efficiently while minimizing code errors. Our SBSFigures demonstrate a strong pre-training effect, making it possible to achieve efficient training with a limited amount of real-world chart data starting from our pre-trained weights.

Kuniaki Saito, Risa Shinoda, Shohei Tanaka et al.

Assessing image-text alignment models such as CLIP is crucial for bridging visual and linguistic representations. Yet existing benchmarks rely on rule-based perturbations or short captions, limiting their ability to measure fine-grained alignment. We introduce AlignBench, a benchmark that provides a new indicator of image-text alignment by evaluating detailed image-caption pairs generated by diverse image-to-text and text-to-image models. Each sentence is annotated for correctness, enabling direct assessment of VLMs as alignment evaluators. Benchmarking a wide range of decoder-based VLMs reveals three key findings: (i) CLIP-based models, even those tailored for compositional reasoning, remain nearly blind; (ii) detectors systematically over-score early sentences; and (iii) they show strong self-preference, favoring their own outputs and harming detection performance. Our project page will be available at https://dahlian00.github.io/AlignBench/.

Dujin Lee, Sojung An, Jungmyung Wi et al.

Universal domain adaptation (UniDA) transfers knowledge from a labeled source domain to an unlabeled target domain, where label spaces may differ and the target domain may contain private classes. Previous UniDA methods primarily focused on visual space alignment but often struggled with visual ambiguities due to content differences, which limited their robustness and generalizability. To overcome this, we introduce a novel approach that leverages the strong \textit{zero-shot capabilities} of recent vision-language foundation models (VLMs) like CLIP, concentrating solely on label space alignment to enhance adaptation stability. CLIP can generate task-specific classifiers based only on label names. However, adapting CLIP to UniDA is challenging because the label space is not fully known in advance. In this study, we first utilize generative vision-language models to identify unknown categories in the target domain. Noise and semantic ambiguities in the discovered labels -- such as those similar to source labels (e.g., synonyms, hypernyms, hyponyms) -- complicate label alignment. To address this, we propose a training-free label-space alignment method for UniDA (\ours). Our method aligns label spaces instead of visual spaces by filtering and refining noisy labels between the domains. We then construct a \textit{universal classifier} that integrates both shared knowledge and target-private class information, thereby improving generalizability under domain shifts. The results reveal that the proposed method considerably outperforms existing UniDA techniques across key DomainBed benchmarks, delivering an average improvement of \textcolor{blue}{+7.9\%}in H-score and \textcolor{blue}{+6.1\%} in H$^3$-score. Furthermore, incorporating self-training further enhances performance and achieves an additional (\textcolor{blue}{+1.6\%}) increment in both H- and H$^3$-scores.

Kuniaki Saito, Ping Hu, Trevor Darrell et al.

Many open-world applications require the detection of novel objects, yet state-of-the-art object detection and instance segmentation networks do not excel at this task. The key issue lies in their assumption that regions without any annotations should be suppressed as negatives, which teaches the model to treat the unannotated objects as background. To address this issue, we propose a simple yet surprisingly powerful data augmentation and training scheme we call Learning to Detect Every Thing (LDET). To avoid suppressing hidden objects, background objects that are visible but unlabeled, we paste annotated objects on a background image sampled from a small region of the original image. Since training solely on such synthetically-augmented images suffers from domain shift, we decouple the training into two parts: 1) training the region classification and regression head on augmented images, and 2)~training the mask heads on original images. In this way, a model does not learn to classify hidden objects as background while generalizing well to real images. LDET leads to significant improvements on many datasets in the open-world instance segmentation task, outperforming baselines on cross-category generalization on COCO, as well as cross-dataset evaluation on UVO and Cityscapes.

Dina Bashkirova, Dan Hendrycks, Donghyun Kim et al.

Progress in machine learning is typically measured by training and testing a model on the same distribution of data, i.e., the same domain. This over-estimates future accuracy on out-of-distribution data. The Visual Domain Adaptation (VisDA) 2021 competition tests models' ability to adapt to novel test distributions and handle distributional shift. We set up unsupervised domain adaptation challenges for image classifiers and will evaluate adaptation to novel viewpoints, backgrounds, modalities and degradation in quality. Our challenge draws on large-scale publicly available datasets but constructs the evaluation across domains, rather that the traditional in-domain bench-marking. Furthermore, we focus on the difficult "universal" setting where, in addition to input distribution drift, methods may encounter missing and/or novel classes in the target dataset. Performance will be measured using a rigorous protocol, comparing to state-of-the-art domain adaptation methods with the help of established metrics. We believe that the competition will encourage further improvement in machine learning methods' ability to handle realistic data in many deployment scenarios.

Kuniaki Saito, Donghyun Kim, Kate Saenko

Semi-supervised learning (SSL) is an effective means to leverage unlabeled data to improve a model's performance. Typical SSL methods like FixMatch assume that labeled and unlabeled data share the same label space. However, in practice, unlabeled data can contain categories unseen in the labeled set, i.e., outliers, which can significantly harm the performance of SSL algorithms. To address this problem, we propose a novel Open-set Semi-Supervised Learning (OSSL) approach called OpenMatch. Learning representations of inliers while rejecting outliers is essential for the success of OSSL. To this end, OpenMatch unifies FixMatch with novelty detection based on one-vs-all (OVA) classifiers. The OVA-classifier outputs the confidence score of a sample being an inlier, providing a threshold to detect outliers. Another key contribution is an open-set soft-consistency regularization loss, which enhances the smoothness of the OVA-classifier with respect to input transformations and greatly improves outlier detection. OpenMatch achieves state-of-the-art performance on three datasets, and even outperforms a fully supervised model in detecting outliers unseen in unlabeled data on CIFAR10.

Kuniaki Saito, Kate Saenko

Universal Domain Adaptation (UNDA) aims to handle both domain-shift and category-shift between two datasets, where the main challenge is to transfer knowledge while rejecting unknown classes which are absent in the labeled source data but present in the unlabeled target data. Existing methods manually set a threshold to reject unknown samples based on validation or a pre-defined ratio of unknown samples, but this strategy is not practical. In this paper, we propose a method to learn the threshold using source samples and to adapt it to the target domain. Our idea is that a minimum inter-class distance in the source domain should be a good threshold to decide between known or unknown in the target. To learn the inter-and intra-class distance, we propose to train a one-vs-all classifier for each class using labeled source data. Then, we adapt the open-set classifier to the target domain by minimizing class entropy. The resulting framework is the simplest of all baselines of UNDA and is insensitive to the value of a hyper-parameter yet outperforms baselines with a large margin.

Donghyun Kim, Kuniaki Saito, Samarth Mishra et al.

Many self-supervised learning (SSL) methods have been successful in learning semantically meaningful visual representations by solving pretext tasks. However, prior work in SSL focuses on tasks like object recognition or detection, which aim to learn object shapes and assume that the features should be invariant to concepts like colors and textures. Thus, these SSL methods perform poorly on downstream tasks where these concepts provide critical information. In this paper, we present an SSL framework that enables us to learn color and texture-aware features without requiring any labels during training. Our approach consists of three self-supervised tasks designed to capture different concepts that are neglected in prior work that we can select from depending on the needs of our downstream tasks. Our tasks include learning to predict color histograms and discriminate shapeless local patches and textures from each instance. We evaluate our approach on fashion compatibility using Polyvore Outfits and In-Shop Clothing Retrieval using Deepfashion, improving upon prior SSL methods by 9.5-16%, and even outperforming some supervised approaches on Polyvore Outfits despite using no labels. We also show that our approach can be used for transfer learning, demonstrating that we can train on one dataset while achieving high performance on a different dataset.

Kuniaki Saito, Kate Saenko, Ming-Yu Liu

Unsupervised image-to-image translation intends to learn a mapping of an image in a given domain to an analogous image in a different domain, without explicit supervision of the mapping. Few-shot unsupervised image-to-image translation further attempts to generalize the model to an unseen domain by leveraging example images of the unseen domain provided at inference time. While remarkably successful, existing few-shot image-to-image translation models find it difficult to preserve the structure of the input image while emulating the appearance of the unseen domain, which we refer to as the content loss problem. This is particularly severe when the poses of the objects in the input and example images are very different. To address the issue, we propose a new few-shot image translation model, COCO-FUNIT, which computes the style embedding of the example images conditioned on the input image and a new module called the constant style bias. Through extensive experimental validations with comparison to the state-of-the-art, our model shows effectiveness in addressing the content loss problem. For code and pretrained models, please check out https://nvlabs.github.io/COCO-FUNIT/ .

Donghyun Kim, Kuniaki Saito, Tae-Hyun Oh et al.

Existing unsupervised domain adaptation methods aim to transfer knowledge from a label-rich source domain to an unlabeled target domain. However, obtaining labels for some source domains may be very expensive, making complete labeling as used in prior work impractical. In this work, we investigate a new domain adaptation scenario with sparsely labeled source data, where only a few examples in the source domain have been labeled, while the target domain is unlabeled. We show that when labeled source examples are limited, existing methods often fail to learn discriminative features applicable for both source and target domains. We propose a novel Cross-Domain Self-supervised (CDS) learning approach for domain adaptation, which learns features that are not only domain-invariant but also class-discriminative. Our self-supervised learning method captures apparent visual similarity with in-domain self-supervision in a domain adaptive manner and performs cross-domain feature matching with across-domain self-supervision. In extensive experiments with three standard benchmark datasets, our method significantly boosts performance of target accuracy in the new target domain with few source labels and is even helpful on classical domain adaptation scenarios.

Donghyun Kim, Kuniaki Saito, Kate Saenko et al.

Existing vision-language methods typically support two languages at a time at most. In this paper, we present a modular approach which can easily be incorporated into existing vision-language methods in order to support many languages. We accomplish this by learning a single shared Multimodal Universal Language Embedding (MULE) which has been visually-semantically aligned across all languages. Then we learn to relate MULE to visual data as if it were a single language. Our method is not architecture specific, unlike prior work which typically learned separate branches for each language, enabling our approach to easily be adapted to many vision-language methods and tasks. Since MULE learns a single language branch in the multimodal model, we can also scale to support many languages, and languages with fewer annotations can take advantage of the good representation learned from other (more abundant) language data. We demonstrate the effectiveness of MULE on the bidirectional image-sentence retrieval task, supporting up to four languages in a single model. In addition, we show that Machine Translation can be used for data augmentation in multilingual learning, which, combined with MULE, improves mean recall by up to 21.9% on a single-language compared to prior work, with the most significant gains seen on languages with relatively few annotations. Our code is publicly available.

Kuniaki Saito, Donghyun Kim, Stan Sclaroff et al.

Contemporary domain adaptation methods are very effective at aligning feature distributions of source and target domains without any target supervision. However, we show that these techniques perform poorly when even a few labeled examples are available in the target. To address this semi-supervised domain adaptation (SSDA) setting, we propose a novel Minimax Entropy (MME) approach that adversarially optimizes an adaptive few-shot model. Our base model consists of a feature encoding network, followed by a classification layer that computes the features' similarity to estimated prototypes (representatives of each class). Adaptation is achieved by alternately maximizing the conditional entropy of unlabeled target data with respect to the classifier and minimizing it with respect to the feature encoder. We empirically demonstrate the superiority of our method over many baselines, including conventional feature alignment and few-shot methods, setting a new state of the art for SSDA.

Toshihiko Matsuura, Kuniaki Saito, Tatsuya Harada

The task of unsupervised domain adaptation is proposed to transfer the knowledge of a label-rich domain (source domain) to a label-scarce domain (target domain). Matching feature distributions between different domains is a widely applied method for the aforementioned task. However, the method does not perform well when classes in the two domains are not identical. Specifically, when the classes of the target correspond to a subset of those of the source, target samples can be incorrectly aligned with the classes that exist only in the source. This problem setting is termed as partial domain adaptation (PDA). In this study, we propose a novel method called Two Weighted Inconsistency-reduced Networks (TWINs) for PDA. We utilize two classification networks to estimate the ratio of the target samples in each class with which a classification loss is weighted to adapt the classes present in the target domain. Furthermore, to extract discriminative features for the target, we propose to minimize the divergence between domains measured by the classifiers' inconsistency on target samples. We empirically demonstrate that reducing the inconsistency between two networks is effective for PDA and that our method outperforms other existing methods with a large margin in several datasets.

Kohei Watanabe, Kuniaki Saito, Yoshitaka Ushiku et al.

Most contemporary robots have depth sensors, and research on semantic segmentation with RGBD images has shown that depth images boost the accuracy of segmentation. Since it is time-consuming to annotate images with semantic labels per pixel, it would be ideal if we could avoid this laborious work by utilizing an existing dataset or a synthetic dataset which we can generate on our own. Robot motions are often tested in a synthetic environment, where multichannel (eg, RGB + depth + instance boundary) images plus their pixel-level semantic labels are available. However, models trained simply on synthetic images tend to demonstrate poor performance on real images. In order to address this, we propose two approaches that can efficiently exploit multichannel inputs combined with an unsupervised domain adaptation (UDA) algorithm. One is a fusion-based approach that uses depth images as inputs. The other is a multitask learning approach that uses depth images as outputs. We demonstrated that the segmentation results were improved by using a multitask learning approach with a post-process and created a benchmark for this task.

Xingchao Peng, Ben Usman, Kuniaki Saito et al.

Unsupervised transfer of object recognition models from synthetic to real data is an important problem with many potential applications. The challenge is how to "adapt" a model trained on simulated images so that it performs well on real-world data without any additional supervision. Unfortunately, current benchmarks for this problem are limited in size and task diversity. In this paper, we present a new large-scale benchmark called Syn2Real, which consists of a synthetic domain rendered from 3D object models and two real-image domains containing the same object categories. We define three related tasks on this benchmark: closed-set object classification, open-set object classification, and object detection. Our evaluation of multiple state-of-the-art methods reveals a large gap in adaptation performance between the easier closed-set classification task and the more difficult open-set and detection tasks. We conclude that developing adaptation methods that work well across all three tasks presents a significant future challenge for syn2real domain transfer.

Kuniaki Saito, Shohei Yamamoto, Yoshitaka Ushiku et al.

Numerous algorithms have been proposed for transferring knowledge from a label-rich domain (source) to a label-scarce domain (target). Almost all of them are proposed for a closed-set scenario, where the source and the target domain completely share the class of their samples. We call the shared class the \doublequote{known class.} However, in practice, when samples in target domain are not labeled, we cannot know whether the domains share the class. A target domain can contain samples of classes that are not shared by the source domain. We call such classes the \doublequote{unknown class} and algorithms that work well in the open set situation are very practical. However, most existing distribution matching methods for domain adaptation do not work well in this setting because unknown target samples should not be aligned with the source. In this paper, we propose a method for an open set domain adaptation scenario which utilizes adversarial training. A classifier is trained to make a boundary between the source and the target samples whereas a generator is trained to make target samples far from the boundary. Thus, we assign two options to the feature generator: aligning them with source known samples or rejecting them as unknown target samples. This approach allows extracting features that separate unknown target samples from known target samples. Our method was extensively evaluated in domain adaptation setting and outperformed other methods with a large margin in most settings.

Kuniaki Saito, Yoshitaka Ushiku, Tatsuya Harada et al.

We present a method for transferring neural representations from label-rich source domains to unlabeled target domains. Recent adversarial methods proposed for this task learn to align features across domains by fooling a special domain critic network. However, a drawback of this approach is that the critic simply labels the generated features as in-domain or not, without considering the boundaries between classes. This can lead to ambiguous features being generated near class boundaries, reducing target classification accuracy. We propose a novel approach, Adversarial Dropout Regularization (ADR), to encourage the generator to output more discriminative features for the target domain. Our key idea is to replace the critic with one that detects non-discriminative features, using dropout on the classifier network. The generator then learns to avoid these areas of the feature space and thus creates better features. We apply our ADR approach to the problem of unsupervised domain adaptation for image classification and semantic segmentation tasks, and demonstrate significant improvement over the state of the art. We also show that our approach can be used to train Generative Adversarial Networks for semi-supervised learning.

Andrew Shin, Leopold Crestel, Hiroharu Kato et al.

Automatic melody generation for pop music has been a long-time aspiration for both AI researchers and musicians. However, learning to generate euphonious melody has turned out to be highly challenging due to a number of factors. Representation of multivariate property of notes has been one of the primary challenges. It is also difficult to remain in the permissible spectrum of musical variety, outside of which would be perceived as a plain random play without auditory pleasantness. Observing the conventional structure of pop music poses further challenges. In this paper, we propose to represent each note and its properties as a unique `word,' thus lessening the prospect of misalignments between the properties, as well as reducing the complexity of learning. We also enforce regularization policies on the range of notes, thus encouraging the generated melody to stay close to what humans would find easy to follow. Furthermore, we generate melody conditioned on song part information, thus replicating the overall structure of a full song. Experimental results demonstrate that our model can generate auditorily pleasant songs that are more indistinguishable from human-written ones than previous models.

Masataka Yamaguchi, Kuniaki Saito, Yoshitaka Ushiku et al.

In this paper, we address the problem of spatio-temporal person retrieval from multiple videos using a natural language query, in which we output a tube (i.e., a sequence of bounding boxes) which encloses the person described by the query. For this problem, we introduce a novel dataset consisting of videos containing people annotated with bounding boxes for each second and with five natural language descriptions. To retrieve the tube of the person described by a given natural language query, we design a model that combines methods for spatio-temporal human detection and multimodal retrieval. We conduct comprehensive experiments to compare a variety of tube and text representations and multimodal retrieval methods, and present a strong baseline in this task as well as demonstrate the efficacy of our tube representation and multimodal feature embedding technique. Finally, we demonstrate the versatility of our model by applying it to two other important tasks.

Kuniaki Saito, Yoshitaka Ushiku, Tatsuya Harada

Deep-layered models trained on a large number of labeled samples boost the accuracy of many tasks. It is important to apply such models to different domains because collecting many labeled samples in various domains is expensive. In unsupervised domain adaptation, one needs to train a classifier that works well on a target domain when provided with labeled source samples and unlabeled target samples. Although many methods aim to match the distributions of source and target samples, simply matching the distribution cannot ensure accuracy on the target domain. To learn discriminative representations for the target domain, we assume that artificially labeling target samples can result in a good representation. Tri-training leverages three classifiers equally to give pseudo-labels to unlabeled samples, but the method does not assume labeling samples generated from a different domain.In this paper, we propose an asymmetric tri-training method for unsupervised domain adaptation, where we assign pseudo-labels to unlabeled samples and train neural networks as if they are true labels. In our work, we use three networks asymmetrically. By asymmetric, we mean that two networks are used to label unlabeled target samples and one network is trained by the samples to obtain target-discriminative representations. We evaluate our method on digit recognition and sentiment analysis datasets. Our proposed method achieves state-of-the-art performance on the benchmark digit recognition datasets of domain adaptation.

Kuniaki Saito, Yusuke Mukuta, Yoshitaka Ushiku et al.

Obtaining common representations from different modalities is important in that they are interchangeable with each other in a classification problem. For example, we can train a classifier on image features in the common representations and apply it to the testing of the text features in the representations. Existing multi-modal representation learning methods mainly aim to extract rich information from paired samples and train a classifier by the corresponding labels; however, collecting paired samples and their labels simultaneously involves high labor costs. Addressing paired modal samples without their labels and single modal data with their labels independently is much easier than addressing labeled multi-modal data. To obtain the common representations under such a situation, we propose to make the distributions over different modalities similar in the learned representations, namely modality-invariant representations. In particular, we propose a novel algorithm for modality-invariant representation learning, named Deep Modality Invariant Adversarial Network (DeMIAN), which utilizes the idea of Domain Adaptation (DA). Using the modality-invariant representations learned by DeMIAN, we achieved better classification accuracy than with the state-of-the-art methods, especially for some benchmark datasets of zero-shot learning.

Kuniaki Saito, Andrew Shin, Yoshitaka Ushiku et al.

Visual question answering (VQA) task not only bridges the gap between images and language, but also requires that specific contents within the image are understood as indicated by linguistic context of the question, in order to generate the accurate answers. Thus, it is critical to build an efficient embedding of images and texts. We implement DualNet, which fully takes advantage of discriminative power of both image and textual features by separately performing two operations. Building an ensemble of DualNet further boosts the performance. Contrary to common belief, our method proved effective in both real images and abstract scenes, in spite of significantly different properties of respective domain. Our method was able to outperform previous state-of-the-art methods in real images category even without explicitly employing attention mechanism, and also outperformed our own state-of-the-art method in abstract scenes category, which recently won the first place in VQA Challenge 2016.