Masanori Suganuma

Yan Zeng, Masanori Suganuma, Takayuki Okatani

This paper studies the problem of inverting the DDIM image generation process to recover latent variables, particularly the initial noise map, from a generated image. Existing methods often struggle with accuracy in this task. We propose a novel hybrid approach that combines direct inversion via gradient descent for the first step, followed by a fixed-point method for subsequent steps. Empirical evaluations across three datasets demonstrate that our method significantly improves the prediction of initial latent variables while achieving superior reconstruction accuracy. Additionally, we introduce a new evaluation, called the self-interpolation test, which assesses the quality of images generated from interpolated points between the true and predicted latent maps, offering deeper insights into performance. Our results reveal that while existing methods perform reasonably well in reconstruction, they consistently fail to accurately predict the initial latent variables, resulting in poor performance on the self-interpolation test. In contrast, our method outperforms all others across all metrics, providing valuable insights into diffusion models and enhancing their applications in image generation and editing.

Xiangyong Lu, Masanori Suganuma, Takayuki Okatani

Computer vision has become increasingly prevalent in solving real-world problems across diverse domains, including smart agriculture, fishery, and livestock management. These applications may not require processing many image frames per second, leading practitioners to use single board computers (SBCs). Although many lightweight networks have been developed for mobile/edge devices, they primarily target smartphones with more powerful processors and not SBCs with the low-end CPUs. This paper introduces a CNN-ViT hybrid network called SBCFormer, which achieves high accuracy and fast computation on such low-end CPUs. The hardware constraints of these CPUs make the Transformer's attention mechanism preferable to convolution. However, using attention on low-end CPUs presents a challenge: high-resolution internal feature maps demand excessive computational resources, but reducing their resolution results in the loss of local image details. SBCFormer introduces an architectural design to address this issue. As a result, SBCFormer achieves the highest trade-off between accuracy and speed on a Raspberry Pi 4 Model B with an ARM-Cortex A72 CPU. For the first time, it achieves an ImageNet-1K top-1 accuracy of around 80% at a speed of 1.0 frame/sec on the SBC. Code is available at https://github.com/xyongLu/SBCFormer.

Jie Zhang, Masanori Suganuma, Takayuki Okatani

Previous works on unsupervised industrial anomaly detection mainly focus on local structural anomalies such as cracks and color contamination. While achieving significantly high detection performance on this kind of anomaly, they are faced with logical anomalies that violate the long-range dependencies such as a normal object placed in the wrong position. In this paper, based on previous knowledge distillation works, we propose to use two students (local and global) to better mimic the teacher's behavior. The local student, which is used in previous studies mainly focuses on structural anomaly detection while the global student pays attention to logical anomalies. To further encourage the global student's learning to capture long-range dependencies, we design the global context condensing block (GCCB) and propose a contextual affinity loss for the student training and anomaly scoring. Experimental results show the proposed method doesn't need cumbersome training techniques and achieves a new state-of-the-art performance on the MVTec LOCO AD dataset.

Van-Quang Nguyen, Masanori Suganuma, Takayuki Okatani

Current state-of-the-art methods for image captioning employ region-based features, as they provide object-level information that is essential to describe the content of images; they are usually extracted by an object detector such as Faster R-CNN. However, they have several issues, such as lack of contextual information, the risk of inaccurate detection, and the high computational cost. The first two could be resolved by additionally using grid-based features. However, how to extract and fuse these two types of features is uncharted. This paper proposes a Transformer-only neural architecture, dubbed GRIT (Grid- and Region-based Image captioning Transformer), that effectively utilizes the two visual features to generate better captions. GRIT replaces the CNN-based detector employed in previous methods with a DETR-based one, making it computationally faster. Moreover, its monolithic design consisting only of Transformers enables end-to-end training of the model. This innovative design and the integration of the dual visual features bring about significant performance improvement. The experimental results on several image captioning benchmarks show that GRIT outperforms previous methods in inference accuracy and speed.

Jie Zhang, Masanori Suganuma, Takayuki Okatani

Recent studies on visual anomaly detection (AD) of industrial objects/textures have achieved quite good performance. They consider an unsupervised setting, specifically the one-class setting, in which we assume the availability of a set of normal (\textit{i.e.}, anomaly-free) images for training. In this paper, we consider a more challenging scenario of unsupervised AD, in which we detect anomalies in a given set of images that might contain both normal and anomalous samples. The setting does not assume the availability of known normal data and thus is completely free from human annotation, which differs from the standard AD considered in recent studies. For clarity, we call the setting blind anomaly detection (BAD). We show that BAD can be converted into a local outlier detection problem and propose a novel method named PatchCluster that can accurately detect image- and pixel-level anomalies. Experimental results show that PatchCluster shows a promising performance without the knowledge of normal data, even comparable to the SOTA methods applied in the one-class setting needing it.

Yusuke Hosoya, Masanori Suganuma, Takayuki Okatani

Open-set object detection (OSOD), a task involving the detection of unknown objects while accurately detecting known objects, has recently gained attention. However, we identify a fundamental issue with the problem formulation employed in current OSOD studies. Inherent to object detection is knowing "what to detect," which contradicts the idea of identifying "unknown" objects. This sets OSOD apart from open-set recognition (OSR). This contradiction complicates a proper evaluation of methods' performance, a fact that previous studies have overlooked. Next, we propose a novel formulation wherein detectors are required to detect both known and unknown classes within specified super-classes of object classes. This new formulation is free from the aforementioned issues and has practical applications. Finally, we design benchmark tests utilizing existing datasets and report the experimental evaluation of existing OSOD methods. The results show that existing methods fail to accurately detect unknown objects due to misclassification of known and unknown classes rather than incorrect bounding box prediction. As a byproduct, we introduce a taxonomy of OSOD, resolving confusion prevalent in the literature. We anticipate that our study will encourage the research community to reconsider OSOD and facilitate progress in the right direction.

Qian Ye, Masanori Suganuma, Jun Xiao et al.

Reconstructing ghosting-free high dynamic range (HDR) images of dynamic scenes from a set of multi-exposure images is a challenging task, especially with large object motion and occlusions, leading to visible artifacts using existing methods. To address this problem, we propose a deep network that tries to learn multi-scale feature flow guided by the regularized loss. It first extracts multi-scale features and then aligns features from non-reference images. After alignment, we use residual channel attention blocks to merge the features from different images. Extensive qualitative and quantitative comparisons show that our approach achieves state-of-the-art performance and produces excellent results where color artifacts and geometric distortions are significantly reduced.

Zhijie Wang, Masanori Suganuma, Takayuki Okatani

Unsupervised domain adaptation (UDA) adapts a model trained on one domain (called source) to a novel domain (called target) using only unlabeled data. Due to its high annotation cost, researchers have developed many UDA methods for semantic segmentation, which assume no labeled sample is available in the target domain. We question the practicality of this assumption for two reasons. First, after training a model with a UDA method, we must somehow verify the model before deployment. Second, UDA methods have at least a few hyper-parameters that need to be determined. The surest solution to these is to evaluate the model using validation data, i.e., a certain amount of labeled target-domain samples. This question about the basic assumption of UDA leads us to rethink UDA from a data-centric point of view. Specifically, we assume we have access to a minimum level of labeled data. Then, we ask how much is necessary to find good hyper-parameters of existing UDA methods. We then consider what if we use the same data for supervised training of the same model, e.g., finetuning. We conducted experiments to answer these questions with popular scenarios, {GTA5, SYNTHIA}$\rightarrow$Cityscapes. We found that i) choosing good hyper-parameters needs only a few labeled images for some UDA methods whereas a lot more for others; and ii) simple finetuning works surprisingly well; it outperforms many UDA methods if only several dozens of labeled images are available.

Han Zou, Masanori Suganuma, Takayuki Okatani

Despite the recent advancement in the study of removing motion blur in an image, it is still hard to deal with strong blurs. While there are limits in removing blurs from a single image, it has more potential to use multiple images, e.g., using an additional image as a reference to deblur a blurry image. A typical setting is deburring an image using a nearby sharp image(s) in a video sequence, as in the studies of video deblurring. This paper proposes a better method to use the information present in a reference image. The method does not need a strong assumption on the reference image. We can utilize an alternative shot of the identical scene, just like in video deblurring, or we can even employ a distinct image from another scene. Our method first matches local patches of the target and reference images and then fuses their features to estimate a sharp image. We employ a patch-based feature matching strategy to solve the difficult problem of matching the blurry image with the sharp reference. Our method can be integrated into pre-existing networks designed for single image deblurring. The experimental results show the effectiveness of the proposed method.

Han Zou, Masanori Suganuma, Takayuki Okatani

Smartphones with multi-camera systems, featuring cameras with varying field-of-views (FoVs), are increasingly common. This variation in FoVs results in content differences across videos, paving the way for an innovative approach to video super-resolution (VSR). This method enhances the VSR performance of lower resolution (LR) videos by leveraging higher resolution reference (Ref) videos. Previous works, which operate on this principle, generally expand on traditional VSR models by combining LR and Ref inputs over time into a unified stream. However, we can expect that better results are obtained by independently aggregating these Ref image sequences temporally. Therefore, we introduce an improved method, RefVSR++, which performs the parallel aggregation of LR and Ref images in the temporal direction, aiming to optimize the use of the available data. RefVSR++ also incorporates improved mechanisms for aligning image features over time, crucial for effective VSR. Our experiments demonstrate that RefVSR++ outperforms previous works by over 1dB in PSNR, setting a new benchmark in the field.

Qian Ye, Masanori Suganuma, Takayuki Okatani

In this paper, we consider the problem in defocus image deblurring. Previous classical methods follow two-steps approaches, i.e., first defocus map estimation and then the non-blind deblurring. In the era of deep learning, some researchers have tried to address these two problems by CNN. However, the simple concatenation of defocus map, which represents the blur level, leads to suboptimal performance. Considering the spatial variant property of the defocus blur and the blur level indicated in the defocus map, we employ the defocus map as conditional guidance to adjust the features from the input blurring images instead of simple concatenation. Then we propose a simple but effective network with spatial modulation based on the defocus map. To achieve this, we design a network consisting of three sub-networks, including the defocus map estimation network, a condition network that encodes the defocus map into condition features, and the defocus deblurring network that performs spatially dynamic modulation based on the condition features. Moreover, the spatially dynamic modulation is based on an affine transform function to adjust the features from the input blurry images. Experimental results show that our method can achieve better quantitative and qualitative evaluation performance than the existing state-of-the-art methods on the commonly used public test datasets.

Korawat Charoenpitaks, Van-Quang Nguyen, Masanori Suganuma et al.

This paper addresses the problem of predicting hazards that drivers may encounter while driving a car. We formulate it as a task of anticipating impending accidents using a single input image captured by car dashcams. Unlike existing approaches to driving hazard prediction that rely on computational simulations or anomaly detection from videos, this study focuses on high-level inference from static images. The problem needs predicting and reasoning about future events based on uncertain observations, which falls under visual abductive reasoning. To enable research in this understudied area, a new dataset named the DHPR (Driving Hazard Prediction and Reasoning) dataset is created. The dataset consists of 15K dashcam images of street scenes, and each image is associated with a tuple containing car speed, a hypothesized hazard description, and visual entities present in the scene. These are annotated by human annotators, who identify risky scenes and provide descriptions of potential accidents that could occur a few seconds later. We present several baseline methods and evaluate their performance on our dataset, identifying remaining issues and discussing future directions. This study contributes to the field by introducing a novel problem formulation and dataset, enabling researchers to explore the potential of multi-modal AI for driving hazard prediction.

Yan Zeng, Masanori Suganuma, Takayuki Okatani

Diffusion models have demonstrated impressive image generation capabilities. Personalized approaches, such as textual inversion and Dreambooth, enhance model individualization using specific images. These methods enable generating images of specific objects based on diverse textual contexts. Our proposed approach aims to retain the model's original knowledge during new information integration, resulting in superior outcomes while necessitating less training time compared to Dreambooth and textual inversion.

Xiangyong Lu, Masanori Suganuma, Takayuki Okatani

In real-world applications of image recognition tasks, such as human pose estimation, cameras often capture objects, like human bodies, at low resolutions. This scenario poses a challenge in extracting and leveraging multi-scale features, which is often essential for precise inference. To address this challenge, we propose a new attention mechanism, named cascaded multi-scale attention (CMSA), tailored for use in CNN-ViT hybrid architectures, to handle low-resolution inputs effectively. The design of CMSA enables the extraction and seamless integration of features across various scales without necessitating the downsampling of the input image or feature maps. This is achieved through a novel combination of grouped multi-head self-attention mechanisms with window-based local attention and cascaded fusion of multi-scale features over different scales. This architecture allows for the effective handling of features across different scales, enhancing the model's ability to perform tasks such as human pose estimation, head pose estimation, and more with low-resolution images. Our experimental results show that the proposed method outperforms existing state-of-the-art methods in these areas with fewer parameters, showcasing its potential for broad application in real-world scenarios where capturing high-resolution images is not feasible. Code is available at https://github.com/xyongLu/CMSA.

Naoya Sogi, Takashi Shibata, Makoto Terao et al.

Result diversification (RD) is a crucial technique in Text-to-Image Retrieval for enhancing the efficiency of a practical application. Conventional methods focus solely on increasing the diversity metric of image appearances. However, the diversity metric and its desired value vary depending on the application, which limits the applications of RD. This paper proposes a novel task called CDR-CA (Contextual Diversity Refinement of Composite Attributes). CDR-CA aims to refine the diversities of multiple attributes, according to the application's context. To address this task, we propose Multi-Source DPPs, a simple yet strong baseline that extends the Determinantal Point Process (DPP) to multi-sources. We model MS-DPP as a single DPP model with a unified similarity matrix based on a manifold representation. We also introduce Tangent Normalization to reflect contexts. Extensive experiments demonstrate the effectiveness of the proposed method. Our code is publicly available at https://github.com/NEC-N-SOGI/msdpp.

Van-Quang Nguyen, Masanori Suganuma, Takayuki Okatani

It has been a primary concern in recent studies of vision and language tasks to design an effective attention mechanism dealing with interactions between the two modalities. The Transformer has recently been extended and applied to several bi-modal tasks, yielding promising results. For visual dialog, it becomes necessary to consider interactions between three or more inputs, i.e., an image, a question, and a dialog history, or even its individual dialog components. In this paper, we present a neural architecture named Light-weight Transformer for Many Inputs (LTMI) that can efficiently deal with all the interactions between multiple such inputs in visual dialog. It has a block structure similar to the Transformer and employs the same design of attention computation, whereas it has only a small number of parameters, yet has sufficient representational power for the purpose. Assuming a standard setting of visual dialog, a layer built upon the proposed attention block has less than one-tenth of parameters as compared with its counterpart, a natural Transformer extension. The experimental results on the VisDial datasets validate the effectiveness of the proposed approach, showing improvements of the best NDCG score on the VisDial v1.0 dataset from 57.59 to 60.92 with a single model, from 64.47 to 66.53 with ensemble models, and even to 74.88 with additional finetuning. Our implementation code is available at https://github.com/davidnvq/visdial.

Korawat Charoenpitaks, Van-Quang Nguyen, Masanori Suganuma et al.

The application of Multi-modal Large Language Models (MLLMs) in Autonomous Driving (AD) faces significant challenges due to their limited training on traffic-specific data and the absence of dedicated benchmarks for spatiotemporal understanding. This study addresses these issues by proposing TB-Bench, a comprehensive benchmark designed to evaluate MLLMs on understanding traffic behaviors across eight perception tasks from ego-centric views. We also introduce vision-language instruction tuning datasets, TB-100k and TB-250k, along with simple yet effective baselines for the tasks. Through extensive experiments, we show that existing MLLMs underperform in these tasks, with even a powerful model like GPT-4o achieving less than 35% accuracy on average. In contrast, when fine-tuned with TB-100k or TB-250k, our baseline models achieve average accuracy up to 85%, significantly enhancing performance on the tasks. Additionally, we demonstrate performance transfer by co-training TB-100k with another traffic dataset, leading to improved performance on the latter. Overall, this study represents a step forward by introducing a comprehensive benchmark, high-quality datasets, and baselines, thus supporting the gradual integration of MLLMs into the perception, prediction, and planning stages of AD.

Lizhi Bai, Chunqi Tian, Jun Yang et al.

3D Gaussian Splatting has emerged as a promising technique for high-quality 3D rendering, leading to increasing interest in integrating 3DGS into realism SLAM systems. However, existing methods face challenges such as Gaussian primitives redundancy, forgetting problem during continuous optimization, and difficulty in initializing primitives in monocular case due to lack of depth information. In order to achieve efficient and photorealistic mapping, we propose RP-SLAM, a 3D Gaussian splatting-based vision SLAM method for monocular and RGB-D cameras. RP-SLAM decouples camera poses estimation from Gaussian primitives optimization and consists of three key components. Firstly, we propose an efficient incremental mapping approach to achieve a compact and accurate representation of the scene through adaptive sampling and Gaussian primitives filtering. Secondly, a dynamic window optimization method is proposed to mitigate the forgetting problem and improve map consistency. Finally, for the monocular case, a monocular keyframe initialization method based on sparse point cloud is proposed to improve the initialization accuracy of Gaussian primitives, which provides a geometric basis for subsequent optimization. The results of numerous experiments demonstrate that RP-SLAM achieves state-of-the-art map rendering accuracy while ensuring real-time performance and model compactness.

Yusuke Hosoya, Masanori Suganuma, Takayuki Okatani

Detecting objects occupying only small areas in an image is difficult, even for humans. Therefore, annotating small-size object instances is hard and thus costly. This study questions common sense by asking the following: is annotating small-size instances worth its cost? We restate it as the following verifiable question: can we detect small-size instances with a detector trained using training data free of small-size instances? We evaluate a method that upscales input images at test time and a method that downscales images at training time. The experiments conducted using the COCO dataset show the following. The first method, together with a remedy to narrow the domain gap between training and test inputs, achieves at least comparable performance to the baseline detector trained using complete training data. Although the method needs to apply the same detector twice to an input image with different scaling, we show that its distillation yields a single-path detector that performs equally well to the same baseline detector. These results point to the necessity of rethinking the annotation of training data for object detection.

Yusuke Hosoya, Masanori Suganuma, Takayuki Okatani

Open-vocabulary object detection (OVD), detecting specific classes of objects using only their linguistic descriptions (e.g., class names) without any image samples, has garnered significant attention. However, in real-world applications, the target class concepts is often hard to describe in text and the only way to specify target objects is to provide their image examples, yet it is often challenging to obtain a good number of samples. Thus, there is a high demand from practitioners for few-shot object detection (FSOD). A natural question arises: Can the benefits of OVD extend to FSOD for object classes that are difficult to describe in text? Compared to traditional methods that learn only predefined classes (referred to in this paper as closed-set object detection, COD), can the extra cost of OVD be justified? To answer these questions, we propose a method to quantify the ``text-describability'' of object detection datasets using the zero-shot image classification accuracy with CLIP. This allows us to categorize various OD datasets with different text-describability and emprically evaluate the FSOD performance of OVD and COD methods within each category. Our findings reveal that: i) there is little difference between OVD and COD for object classes with low text-describability under equal conditions in OD pretraining; and ii) although OVD can learn from more diverse data than OD-specific data, thereby increasing the volume of training data, it can be counterproductive for classes with low-text-describability. These findings provide practitioners with valuable guidance amidst the recent advancements of OVD methods.

Zhijie Wang, Masanori Suganuma, Takayuki Okatani

This study is concerned with few-shot segmentation, i.e., segmenting the region of an unseen object class in a query image, given support image(s) of its instances. The current methods rely on the pretrained CNN features of the support and query images. The key to good performance depends on the proper fusion of their mid-level and high-level features; the former contains shape-oriented information, while the latter has class-oriented information. Current state-of-the-art methods follow the approach of Tian et al., which gives the mid-level features the primary role and the high-level features the secondary role. In this paper, we reinterpret this widely employed approach by redifining the roles of the multi-level features; we swap the primary and secondary roles. Specifically, we regard that the current methods improve the initial estimate generated from the high-level features using the mid-level features. This reinterpretation suggests a new application of the current methods: to apply the same network multiple times to iteratively update the estimate of the object's region, starting from its initial estimate. Our experiments show that this method is effective and has updated the previous state-of-the-art on COCO-20$^i$ in the 1-shot and 5-shot settings and on PASCAL-5$^i$ in the 1-shot setting.

Zhijie Wang, Xing Liu, Masanori Suganuma et al.

Semantic segmentation requires a lot of training data, which necessitates costly annotation. There have been many studies on unsupervised domain adaptation (UDA) from one domain to another, e.g., from computer graphics to real images. However, there is still a gap in accuracy between UDA and supervised training on native domain data. It is arguably attributable to class-level misalignment between the source and target domain data. To cope with this, we propose a method that applies adversarial training to align two feature distributions in the target domain. It uses a self-training framework to split the image into two regions (i.e., trusted and untrusted), which form two distributions to align in the feature space. We term this approach cross-region adaptation (CRA) to distinguish from the previous methods of aligning different domain distributions, which we call cross-domain adaptation (CDA). CRA can be applied after any CDA method. Experimental results show that this always improves the accuracy of the combined CDA method, having updated the state-of-the-art.

Wenzheng Song, Masanori Suganuma, Xing Liu et al.

This paper considers matching images of low-light scenes, aiming to widen the frontier of SfM and visual SLAM applications. Recent image sensors can record the brightness of scenes with more than eight-bit precision, available in their RAW-format image. We are interested in making full use of such high-precision information to match extremely low-light scene images that conventional methods cannot handle. For extreme low-light scenes, even if some of their brightness information exists in the RAW format images' low bits, the standard raw image processing on cameras fails to utilize them properly. As was recently shown by Chen et al., CNNs can learn to produce images with a natural appearance from such RAW-format images. To consider if and how well we can utilize such information stored in RAW-format images for image matching, we have created a new dataset named MID (matching in the dark). Using it, we experimentally evaluated combinations of eight image-enhancing methods and eleven image matching methods consisting of classical/neural local descriptors and classical/neural initial point-matching methods. The results show the advantage of using the RAW-format images and the strengths and weaknesses of the above component methods. They also imply there is room for further research.

Van-Quang Nguyen, Masanori Suganuma, Takayuki Okatani

There is a growing interest in the community in making an embodied AI agent perform a complicated task while interacting with an environment following natural language directives. Recent studies have tackled the problem using ALFRED, a well-designed dataset for the task, but achieved only very low accuracy. This paper proposes a new method, which outperforms the previous methods by a large margin. It is based on a combination of several new ideas. One is a two-stage interpretation of the provided instructions. The method first selects and interprets an instruction without using visual information, yielding a tentative action sequence prediction. It then integrates the prediction with the visual information etc., yielding the final prediction of an action and an object. As the object's class to interact is identified in the first stage, it can accurately select the correct object from the input image. Moreover, our method considers multiple egocentric views of the environment and extracts essential information by applying hierarchical attention conditioned on the current instruction. This contributes to the accurate prediction of actions for navigation. A preliminary version of the method won the ALFRED Challenge 2020. The current version achieves the unseen environment's success rate of 4.45% with a single view, which is further improved to 8.37% with multiple views.

Rito Murase, Masanori Suganuma, Takayuki Okatani

Convolution is an equivariant operation, and image position does not affect its result. A recent study shows that the zero-padding employed in convolutional layers of CNNs provides position information to the CNNs. The study further claims that the position information enables accurate inference for several tasks, such as object recognition, segmentation, etc. However, there is a technical issue with the design of the experiments of the study, and thus the correctness of the claim is yet to be verified. Moreover, the absolute image position may not be essential for the segmentation of natural images, in which target objects will appear at any image position. In this study, we investigate how positional information is and can be utilized for segmentation tasks. Toward this end, we consider {\em positional encoding} (PE) that adds channels embedding image position to the input images and compare PE with several padding methods. Considering the above nature of natural images, we choose medical image segmentation tasks, in which the absolute position appears to be relatively important, as the same organs (of different patients) are captured in similar sizes and positions. We draw a mixed conclusion from the experimental results; the positional encoding certainly works in some cases, but the absolute image position may not be so important for segmentation tasks as we think.

Yusuke Hosoya, Masanori Suganuma, Takayuki Okatani

The employment of convolutional neural networks has led to significant performance improvement on the task of object detection. However, when applying existing detectors to continuous frames in a video, we often encounter momentary miss-detection of objects, that is, objects are undetected exceptionally at a few frames, although they are correctly detected at all other frames. In this paper, we analyze the mechanism of how such miss-detection occurs. For the most popular class of detectors that are based on anchor boxes, we show the followings: i) besides apparent causes such as motion blur, occlusions, background clutters, etc., the majority of remaining miss-detection can be explained by an improper behavior of the detectors at boundaries of the anchor boxes; and ii) this can be rectified by improving the way of choosing positive samples from candidate anchor boxes when training the detectors.

Xing Liu, Masanori Suganuma, Xiyang Luo et al.

The employment of convolutional neural networks has achieved unprecedented performance in the task of image restoration for a variety of degradation factors. However, high-performance networks have been specifically designed for a single degradation factor. In this paper, we tackle a harder problem, restoring a clean image from its degraded version with an unknown degradation factor, subject to the condition that it is one of the known factors. Toward this end, we design a network having multiple pairs of input and output branches and use it in a recurrent fashion such that a different branch pair is used at each of the recurrent paths. We reinforce the shared part of the network with improved components so that it can handle different degradation factors. We also propose a two-step training method for the network, which consists of multi-task learning and finetuning. The experimental results show that the proposed network yields at least comparable or sometimes even better performance on four degradation factors as compared with the best dedicated network for each of the four. We also test it on a further harder task where the input image contains multiple degradation factors that are mixed with unknown mixture ratios, showing that it achieves better performance than the previous state-of-the-art method designed for the task.

Engkarat Techapanurak, Masanori Suganuma, Takayuki Okatani

The ability to detect out-of-distribution (OOD) samples is vital to secure the reliability of deep neural networks in real-world applications. Considering the nature of OOD samples, detection methods should not have hyperparameters that need to be tuned depending on incoming OOD samples. However, most of the recently proposed methods do not meet this requirement, leading to compromised performance in real-world applications. In this paper, we propose a simple, hyperparameter-free method based on softmax of scaled cosine similarity. It resembles the approach employed by modern metric learning methods, but it differs in details; the differences are essential to achieve high detection performance. We show through experiments that our method outperforms the existing methods on the evaluation test recently proposed by Shafaei et al., which takes the above issue of hyperparameter dependency into account. We also show that it achieves at least comparable performance to other methods on the conventional test, where their hyperparameters are chosen using explicit OOD samples. Furthermore, it is computationally more efficient than most of the previous methods, since it needs only a single forward pass.

Xing Liu, Masanori Suganuma, Zhun Sun et al.

In this paper, we study design of deep neural networks for tasks of image restoration. We propose a novel style of residual connections dubbed "dual residual connection", which exploits the potential of paired operations, e.g., up- and down-sampling or convolution with large- and small-size kernels. We design a modular block implementing this connection style; it is equipped with two containers to which arbitrary paired operations are inserted. Adopting the "unraveled" view of the residual networks proposed by Veit et al., we point out that a stack of the proposed modular blocks allows the first operation in a block interact with the second operation in any subsequent blocks. Specifying the two operations in each of the stacked blocks, we build a complete network for each individual task of image restoration. We experimentally evaluate the proposed approach on five image restoration tasks using nine datasets. The results show that the proposed networks with properly chosen paired operations outperform previous methods on almost all of the tasks and datasets.

Masanori Suganuma, Xing Liu, Takayuki Okatani

Many studies have been conducted so far on image restoration, the problem of restoring a clean image from its distorted version. There are many different types of distortion which affect image quality. Previous studies have focused on single types of distortion, proposing methods for removing them. However, image quality degrades due to multiple factors in the real world. Thus, depending on applications, e.g., vision for autonomous cars or surveillance cameras, we need to be able to deal with multiple combined distortions with unknown mixture ratios. For this purpose, we propose a simple yet effective layer architecture of neural networks. It performs multiple operations in parallel, which are weighted by an attention mechanism to enable selection of proper operations depending on the input. The layer can be stacked to form a deep network, which is differentiable and thus can be trained in an end-to-end fashion by gradient descent. The experimental results show that the proposed method works better than previous methods by a good margin on tasks of restoring images with multiple combined distortions.

Masanori Suganuma, Mete Ozay, Takayuki Okatani

Researchers have applied deep neural networks to image restoration tasks, in which they proposed various network architectures, loss functions, and training methods. In particular, adversarial training, which is employed in recent studies, seems to be a key ingredient to success. In this paper, we show that simple convolutional autoencoders (CAEs) built upon only standard network components, i.e., convolutional layers and skip connections, can outperform the state-of-the-art methods which employ adversarial training and sophisticated loss functions. The secret is to employ an evolutionary algorithm to automatically search for good architectures. Training optimized CAEs by minimizing the $\ell_2$ loss between reconstructed images and their ground truths using the ADAM optimizer is all we need. Our experimental results show that this approach achieves 27.8 dB peak signal to noise ratio (PSNR) on the CelebA dataset and 40.4 dB on the SVHN dataset, compared to 22.8 dB and 33.0 dB provided by the former state-of-the-art methods, respectively.

Masanori Suganuma, Shinichi Shirakawa, Tomoharu Nagao

The convolutional neural network (CNN), which is one of the deep learning models, has seen much success in a variety of computer vision tasks. However, designing CNN architectures still requires expert knowledge and a lot of trial and error. In this paper, we attempt to automatically construct CNN architectures for an image classification task based on Cartesian genetic programming (CGP). In our method, we adopt highly functional modules, such as convolutional blocks and tensor concatenation, as the node functions in CGP. The CNN structure and connectivity represented by the CGP encoding method are optimized to maximize the validation accuracy. To evaluate the proposed method, we constructed a CNN architecture for the image classification task with the CIFAR-10 dataset. The experimental result shows that the proposed method can be used to automatically find the competitive CNN architecture compared with state-of-the-art models.