Rei Kawakami

Shingo Yashima, Teppei Suzuki, Kohta Ishikawa et al.

Ensembles of deep neural networks demonstrate improved performance over single models. For enhancing the diversity of ensemble members while keeping their performance, particle-based inference methods offer a promising approach from a Bayesian perspective. However, the best way to apply these methods to neural networks is still unclear: seeking samples from the weight-space posterior suffers from inefficiency due to the over-parameterization issues, while seeking samples directly from the function-space posterior often results in serious underfitting. In this study, we propose optimizing particles in the feature space where the activation of a specific intermediate layer lies to address the above-mentioned difficulties. Our method encourages each member to capture distinct features, which is expected to improve ensemble prediction robustness. Extensive evaluation on real-world datasets shows that our model significantly outperforms the gold-standard Deep Ensembles on various metrics, including accuracy, calibration, and robustness. Code is available at https://github.com/DensoITLab/featurePI .

Ikuro Sato, Ryota Yamada, Masayuki Tanaka et al.

It has been intensively investigated that the local shape, especially flatness, of the loss landscape near a minimum plays an important role for generalization of deep models. We developed a training algorithm called PoF: Post-Training of Feature Extractor that updates the feature extractor part of an already-trained deep model to search a flatter minimum. The characteristics are two-fold: 1) Feature extractor is trained under parameter perturbations in the higher-layer parameter space, based on observations that suggest flattening higher-layer parameter space, and 2) the perturbation range is determined in a data-driven manner aiming to reduce a part of test loss caused by the positive loss curvature. We provide a theoretical analysis that shows the proposed algorithm implicitly reduces the target Hessian components as well as the loss. Experimental results show that PoF improved model performance against baseline methods on both CIFAR-10 and CIFAR-100 datasets for only 10-epoch post-training, and on SVHN dataset for 50-epoch post-training. Source code is available at: \url{https://github.com/DensoITLab/PoF-v1

Tatsukichi Shibuya, Nakamasa Inoue, Rei Kawakami et al.

Target Propagation (TP) is a biologically more plausible algorithm than the error backpropagation (BP) to train deep networks, and improving practicality of TP is an open issue. TP methods require the feedforward and feedback networks to form layer-wise autoencoders for propagating the target values generated at the output layer. However, this causes certain drawbacks; e.g., careful hyperparameter tuning is required to synchronize the feedforward and feedback training, and frequent updates of the feedback path are usually required than that of the feedforward path. Learning of the feedforward and feedback networks is sufficient to make TP methods capable of training, but is having these layer-wise autoencoders a necessary condition for TP to work? We answer this question by presenting Fixed-Weight Difference Target Propagation (FW-DTP) that keeps the feedback weights constant during training. We confirmed that this simple method, which naturally resolves the abovementioned problems of TP, can still deliver informative target values to hidden layers for a given task; indeed, FW-DTP consistently achieves higher test performance than a baseline, the Difference Target Propagation (DTP), on four classification datasets. We also present a novel propagation architecture that explains the exact form of the feedback function of DTP to analyze FW-DTP.

Nakamasa Inoue, Shinta Otake, Takumi Hirose et al.

Self-supervised learning has emerged as a key approach for learning generic representations from speech data. Despite promising results in downstream tasks such as speech recognition, speaker verification, and emotion recognition, a significant number of parameters is required, which makes fine-tuning for each task memory-inefficient. To address this limitation, we introduce ELP-adapter tuning, a novel method for parameter-efficient fine-tuning using three types of adapter, namely encoder adapters (E-adapters), layer adapters (L-adapters), and a prompt adapter (P-adapter). The E-adapters are integrated into transformer-based encoder layers and help to learn fine-grained speech representations that are effective for speech recognition. The L-adapters create paths from each encoder layer to the downstream head and help to extract non-linguistic features from lower encoder layers that are effective for speaker verification and emotion recognition. The P-adapter appends pseudo features to CNN features to further improve effectiveness and efficiency. With these adapters, models can be quickly adapted to various speech processing tasks. Our evaluation across four downstream tasks using five backbone models demonstrated the effectiveness of the proposed method. With the WavLM backbone, its performance was comparable to or better than that of full fine-tuning on all tasks while requiring 90% fewer learnable parameters.

Aoyu Li, Ikuro Sato, Kohta Ishikawa et al.

Among various supervised deep metric learning methods proxy-based approaches have achieved high retrieval accuracies. Proxies, which are class-representative points in an embedding space, receive updates based on proxy-sample similarities in a similar manner to sample representations. In existing methods, a relatively small number of samples can produce large gradient magnitudes (ie, hard samples), and a relatively large number of samples can produce small gradient magnitudes (ie, easy samples); these can play a major part in updates. Assuming that acquiring too much sensitivity to such extreme sets of samples would deteriorate the generalizability of a method, we propose a novel proxy-based method called Informative Sample-Aware Proxy (Proxy-ISA), which directly modifies a gradient weighting factor for each sample using a scheduled threshold function, so that the model is more sensitive to the informative samples. Extensive experiments on the CUB-200-2011, Cars-196, Stanford Online Products and In-shop Clothes Retrieval datasets demonstrate the superiority of Proxy-ISA compared with the state-of-the-art methods.

Masahiro Kada, Ryota Yoshihashi, Satoshi Ikehata et al.

Recent progress in deep learning has been driven by increasingly large-scale models, but the resulting computational cost has become a critical bottleneck. Sparse Mixture of Experts (MoE) offers an effective solution by activating only a small subset of experts for each input, achieving high scalability without sacrificing inference speed. Although effective, sparse MoE training exhibits characteristic optimization difficulties. Because the router receives informative gradients only through the experts selected in the forward pass, it suffers from gradient blocking and obtains little information from unselected routes. This limited, highly localized feedback makes it difficult for the router to learn appropriate expert-selection scores and often leads to unstable routing dynamics, such as fluctuating expert assignments during training. To address this issue, we propose TGR-MoE: Teacher-Guided Routing for Sparse Vision Mixture-of-Experts, a simple yet effective method that stabilizes router learning using supervision derived from a pretrained dense teacher model. TGR-MoE constructs a teacher router from the teacher's intermediate representations and uses its routing outputs as pseudo-supervision for the student router, suppressing frequent routing fluctuations during training and enabling knowledge-guided expert selection from the early stages of training. Extensive experiments on ImageNet-1K and CIFAR-100 demonstrate that TGR consistently improves both accuracy and routing consistency, while maintaining stable training even under highly sparse configurations.

Ryota Yoshihashi, Masahiro Kada, Satoshi Ikehata et al.

Many image understanding tasks involve identifying what is present and where it appears. However, tasks that address where, such as object discovery, detection, and segmentation, are often considerably more complex than image classification, which primarily focuses on what. One possible reason is that classification-oriented backbones tend to emphasize semantic information about what, while implicitly entangling or suppressing information about where. In this work, we focus on an inductive bias termed what-where separation, which encourages models to represent object appearance and spatial location in a decomposed manner. To incorporate this bias throughout an attentive backbone in the style of Vision Transformer (ViT), we propose the What-Where Transformer (WWT). Our method introduces two key novel designs: (1) it treats tokens as representations of what and attention maps as representations of where, and processes them in concurrent feed-forward modules via a multi-stream, slot-based architecture; (2) it reuses both the final-layer tokens and attention maps for downstream tasks, and directly exposes them to gradients derived from task losses, thereby facilitating more effective and explicit learning of localization. We demonstrate that even under standard single-label classification-based supervision on ImageNet, WWT exhibits emergent multiple object discovery directly from raw attention maps, rather than via additional postprocessing such as token clustering. Furthermore, WWT achieves superior performance compared to ViT-based methods on zero-shot object discovery and weakly supervised semantic segmentation, and it is transferable to various localization setups with minimal modifications. Code will be published after acceptance.

Eunho Lee, Youngbae Hwang, Rei Kawakami

All-in-one image restoration seeks to recover clean images from inputs affected by diverse and unknown degradations using a unified framework. Recent methods have shown strong performance by identifying degradation characteristics to guide the restoration process. However, many of them treat degradations as discrete categories, which limits their ability to model the continuous relational structure that arises in composite degradations. To address this issue, we propose a multimodal large language model (MLLM)-guided image restoration framework that exploits multimodal embeddings as guidance for low-level restoration. Specifically, MLLM-derived features are injected into an encoder-decoder architecture through an MLLM-guided fusion block (MGFB) to enhance degradation-aware representations. In addition, we incorporate a mixture-of-frequency-experts (MoFE) module that adaptively combines frequency experts using MLLM-guided contextual cues. To further improve expert routing, we design an MLLM-guided router with a relational alignment loss that encourages routing patterns consistent with the embedding-space relationships of degraded inputs. Extensive experiments on multiple benchmarks show that the proposed method achieves strong performance across diverse restoration settings and establishes a new state of the art on the challenging CDD11 dataset, outperforming previous methods by up to 1.35 dB.

Ryota Yoshihashi, Wen Shao, Rei Kawakami et al.

Open-set classification is a problem of handling `unknown' classes that are not contained in the training dataset, whereas traditional classifiers assume that only known classes appear in the test environment. Existing open-set classifiers rely on deep networks trained in a supervised manner on known classes in the training set; this causes specialization of learned representations to known classes and makes it hard to distinguish unknowns from knowns. In contrast, we train networks for joint classification and reconstruction of input data. This enhances the learned representation so as to preserve information useful for separating unknowns from knowns, as well as to discriminate classes of knowns. Our novel Classification-Reconstruction learning for Open-Set Recognition (CROSR) utilizes latent representations for reconstruction and enables robust unknown detection without harming the known-class classification accuracy. Extensive experiments reveal that the proposed method outperforms existing deep open-set classifiers in multiple standard datasets and is robust to diverse outliers. The code is available in https://nae-lab.org/~rei/research/crosr/.

Ryo Moriai, Nakamasa Inoue, Masayuki Tanaka et al.

Modern Hopfield networks (MHNs) have recently gained significant attention in the field of artificial intelligence because they can store and retrieve a large set of patterns with an exponentially large memory capacity. A MHN is generally a dynamical system defined with Lagrangians of memory and feature neurons, where memories associated with in-distribution (ID) samples are represented by attractors in the feature space. One major problem in existing MHNs lies in managing out-of-distribution (OOD) samples because it was originally assumed that all samples are ID samples. To address this, we propose the rectified Lagrangian (RegLag), a new Lagrangian for memory neurons that explicitly incorporates an attractor for OOD samples in the dynamical system of MHNs. RecLag creates a trivial point attractor for any interaction matrix, enabling OOD detection by identifying samples that fall into this attractor as OOD. The interaction matrix is optimized so that the probability densities can be estimated to identify ID/OOD. We demonstrate the effectiveness of RecLag-based MHNs compared to energy-based OOD detection methods, including those using state-of-the-art Hopfield energies, across nine image datasets.

Yusuke Sekikawa, Chingwei Hsu, Satoshi Ikehata et al.

We propose Gumbel-NeRF, a mixture-of-expert (MoE) neural radiance fields (NeRF) model with a hindsight expert selection mechanism for synthesizing novel views of unseen objects. Previous studies have shown that the MoE structure provides high-quality representations of a given large-scale scene consisting of many objects. However, we observe that such a MoE NeRF model often produces low-quality representations in the vicinity of experts' boundaries when applied to the task of novel view synthesis of an unseen object from one/few-shot input. We find that this deterioration is primarily caused by the foresight expert selection mechanism, which may leave an unnatural discontinuity in the object shape near the experts' boundaries. Gumbel-NeRF adopts a hindsight expert selection mechanism, which guarantees continuity in the density field even near the experts' boundaries. Experiments using the SRN cars dataset demonstrate the superiority of Gumbel-NeRF over the baselines in terms of various image quality metrics.

King-Man Tam, Satoshi Ikehata, Yuta Asano et al.

Universal Photometric Stereo is a promising approach for recovering surface normals without strict lighting assumptions. However, it struggles when multi-illumination cues are unreliable, such as under biased lighting or in shadows or self-occluded regions of complex in-the-wild scenes. We propose GeoUniPS, a universal photometric stereo network that integrates synthetic supervision with high-level geometric priors from large-scale 3D reconstruction models pretrained on massive in-the-wild data. Our key insight is that these 3D reconstruction models serve as visual-geometry foundation models, inherently encoding rich geometric knowledge of real scenes. To leverage this, we design a Light-Geometry Dual-Branch Encoder that extracts both multi-illumination cues and geometric priors from the frozen 3D reconstruction model. We also address the limitations of the conventional orthographic projection assumption by introducing the PS-Perp dataset with realistic perspective projection to enable learning of spatially varying view directions. Extensive experiments demonstrate that GeoUniPS delivers state-of-the-arts performance across multiple datasets, both quantitatively and qualitatively, especially in the complex in-the-wild scenes.

Zhaoyi An, Rei Kawakami

Large language models, with their strong reasoning ability and rich knowledge, have brought revolution to many tasks of AI, but their impact on sign language generation remains limited due to its complexity and unique rules. In this paper, we propose TEAch Me Sign (TEAM-Sign), treating sign language as another natural language. By fine-tuning an LLM, we enable it to learn the correspondence between text and sign language, and facilitate generation. Considering the differences between sign and spoken language, we employ a stepwise prompting strategy to extract the inherent sign language knowledge within the LLM, thereby supporting the learning and generation process. Experimental results on How2Sign and Phoenix14T datasets demonstrate that our approach effectively leverages both the sign language knowledge and reasoning capabilities of LLM to align the different distribution and grammatical rules between sign and spoken language.

Edward Effendy, Kuan-Wei Tseng, Rei Kawakami

Accepted in the ICIP 2025 We present a novel transformer-based framework for whole-body grasping that addresses both pose generation and motion infilling, enabling realistic and stable object interactions. Our pipeline comprises three stages: Grasp Pose Generation for full-body grasp generation, Temporal Infilling for smooth motion continuity, and a LiftUp Transformer that refines downsampled joints back to high-resolution markers. To overcome the scarcity of hand-object interaction data, we introduce a data-efficient Generalized Pretraining stage on large, diverse motion datasets, yielding robust spatio-temporal representations transferable to grasping tasks. Experiments on the GRAB dataset show that our method outperforms state-of-the-art baselines in terms of coherence, stability, and visual realism. The modular design also supports easy adaptation to other human-motion applications.

Daichi Tanaka, Takumi Karasawa, Shu Takenouchi et al.

Recycling steel scrap can reduce carbon dioxide (CO2) emissions from the steel industry. However, a significant challenge in steel scrap recycling is the inclusion of impurities other than steel. To address this issue, we propose vision-language-model-based anomaly detection where a model is finetuned in a supervised manner, enabling it to handle niche objects effectively. This model enables automated detection of anomalies at a fine-grained level within steel scrap. Specifically, we finetune the image encoder, equipped with multi-scale mechanism and text prompts aligned with both normal and anomaly images. The finetuning process trains these modules using a multiclass classification as the supervision.

Masaru Yajima, Kei Ota, Asako Kanezaki et al.

Achieving zero-shot peg insertion, where inserting an arbitrary peg into an unseen hole without task-specific training, remains a fundamental challenge in robotics. This task demands a highly generalizable perception system capable of detecting potential holes, selecting the correct mating hole from multiple candidates, estimating its precise pose, and executing insertion despite uncertainties. While learning-based methods have been applied to peg insertion, they often fail to generalize beyond the specific peg-hole pairs encountered during training. Recent advancements in Vision-Language Models (VLMs) offer a promising alternative, leveraging large-scale datasets to enable robust generalization across diverse tasks. Inspired by their success, we introduce a novel zero-shot peg insertion framework that utilizes a VLM to identify mating holes and estimate their poses without prior knowledge of their geometry. Extensive experiments demonstrate that our method achieves 90.2% accuracy, significantly outperforming baselines in identifying the correct mating hole across a wide range of previously unseen peg-hole pairs, including 3D-printed objects, toy puzzles, and industrial connectors. Furthermore, we validate the effectiveness of our approach in a real-world connector insertion task on a backpanel of a PC, where our system successfully detects holes, identifies the correct mating hole, estimates its pose, and completes the insertion with a success rate of 88.3%. These results highlight the potential of VLM-driven zero-shot reasoning for enabling robust and generalizable robotic assembly.

Kanoko Goto, Takumi Karasawa, Takumi Hirose et al.

Small object detection aims to localize and classify small objects within images. With recent advances in large-scale vision-language pretraining, finetuning pretrained object detection models has emerged as a promising approach. However, finetuning large models is computationally and memory expensive. To address this issue, this paper introduces multi-point positional insertion (MPI) tuning, a parameter-efficient finetuning (PEFT) method for small object detection. Specifically, MPI incorporates multiple positional embeddings into a frozen pretrained model, enabling the efficient detection of small objects by providing precise positional information to latent features. Through experiments, we demonstrated the effectiveness of the proposed method on the SODA-D dataset. MPI performed comparably to conventional PEFT methods, including CoOp and VPT, while significantly reducing the number of parameters that need to be tuned.

Ryota Yoshihashi, Rei Kawakami, Shaodi You et al.

Detecting tiny objects in a high-resolution video is challenging because the visual information is little and unreliable. Specifically, the challenge includes very low resolution of the objects, MPEG artifacts due to compression and a large searching area with many hard negatives. Tracking is equally difficult because of the unreliable appearance, and the unreliable motion estimation. Luckily, we found that by combining this two challenging tasks together, there will be mutual benefits. Following the idea, in this paper, we present a neural network model called the Recurrent Correlational Network, where detection and tracking are jointly performed over a multi-frame representation learned through a single, trainable, and end-to-end network. The framework exploits a convolutional long short-term memory network for learning informative appearance changes for detection, while the learned representation is shared in tracking for enhancing its performance. In experiments with datasets containing images of scenes with small flying objects, such as birds and unmanned aerial vehicles, the proposed method yielded consistent improvements in detection performance over deep single-frame detectors and existing motion-based detectors. Furthermore, our network performs as well as state-of-the-art generic object trackers when it was evaluated as a tracker on a bird image dataset.

Kenta Moriwaki, Ryota Yoshihashi, Rei Kawakami et al.

This paper tackles high-dynamic-range (HDR) image reconstruction given only a single low-dynamic-range (LDR) image as input. While the existing methods focus on minimizing the mean-squared-error (MSE) between the target and reconstructed images, we minimize a hybrid loss that consists of perceptual and adversarial losses in addition to HDR-reconstruction loss. The reconstruction loss instead of MSE is more suitable for HDR since it puts more weight on both over- and under- exposed areas. It makes the reconstruction faithful to the input. Perceptual loss enables the networks to utilize knowledge about objects and image structure for recovering the intensity gradients of saturated and grossly quantized areas. Adversarial loss helps to select the most plausible appearance from multiple solutions. The hybrid loss that combines all the three losses is calculated in logarithmic space of image intensity so that the outputs retain a large dynamic range and meanwhile the learning becomes tractable. Comparative experiments conducted with other state-of-the-art methods demonstrated that our method produces a leap in image quality.

Seiichiro Fukuda, Ryota Yoshihashi, Rei Kawakami et al.

Multi-task learning improves generalization performance by sharing knowledge among related tasks. Existing models are for task combinations annotated on the same dataset, while there are cases where multiple datasets are available for each task. How to utilize knowledge of successful single-task CNNs that are trained on each dataset has been explored less than multi-task learning with a single dataset. We propose a cross-connected CNN, a new architecture that connects single-task CNNs through convolutional layers, which transfer useful information for the counterpart. We evaluated our proposed architecture on a combination of detection and segmentation using two datasets. Experiments on pedestrians show our CNN achieved a higher detection performance compared to baseline CNNs, while maintaining high quality for segmentation. It is the first known attempt to tackle multi-task learning with different training datasets between detection and segmentation. Experiments with wild birds demonstrate how our CNN learns general representations from limited datasets.

Ryota Yoshihashi, Tu Tuan Trinh, Rei Kawakami et al.

While generic object detection has achieved large improvements with rich feature hierarchies from deep nets, detecting small objects with poor visual cues remains challenging. Motion cues from multiple frames may be more informative for detecting such hard-to-distinguish objects in each frame. However, how to encode discriminative motion patterns, such as deformations and pose changes that characterize objects, has remained an open question. To learn them and thereby realize small object detection, we present a neural model called the Recurrent Correlational Network, where detection and tracking are jointly performed over a multi-frame representation learned through a single, trainable, and end-to-end network. A convolutional long short-term memory network is utilized for learning informative appearance change for detection, while learned representation is shared in tracking for enhancing its performance. In experiments with datasets containing images of scenes with small flying objects, such as birds and unmanned aerial vehicles, the proposed method yielded consistent improvements in detection performance over deep single-frame detectors and existing motion-based detectors. Furthermore, our network performs as well as state-of-the-art generic object trackers when it was evaluated as a tracker on the bird dataset.

Shaodi You, Robby T. Tan, Rei Kawakami et al.

This paper introduces depth estimation from water drops. The key idea is that a single water drop adhered to window glass is totally transparent and convex, and thus optically acts like a fisheye lens. If we have more than one water drop in a single image, then through each of them we can see the environment with different view points, similar to stereo. To realize this idea, we need to rectify every water drop imagery to make radially distorted planar surfaces look flat. For this rectification, we consider two physical properties of water drops: (1) A static water drop has constant volume, and its geometric convex shape is determined by the balance between the tension force and gravity. This implies that the 3D geometric shape can be obtained by minimizing the overall potential energy, which is the sum of the tension energy and the gravitational potential energy. (2) The imagery inside a water-drop is determined by the water-drop 3D shape and total reflection at the boundary. This total reflection generates a dark band commonly observed in any adherent water drops. Hence, once the 3D shape of water drops are recovered, we can rectify the water drop images through backward raytracing. Subsequently, we can compute depth using stereo. In addition to depth estimation, we can also apply image refocusing. Experiments on real images and a quantitative evaluation show the effectiveness of our proposed method. To our best knowledge, never before have adherent water drops been used to estimate depth.