Sanghyun Woo

Sanghyun Woo, Shoubhik Debnath, Ronghang Hu et al.

Driven by improved architectures and better representation learning frameworks, the field of visual recognition has enjoyed rapid modernization and performance boost in the early 2020s. For example, modern ConvNets, represented by ConvNeXt, have demonstrated strong performance in various scenarios. While these models were originally designed for supervised learning with ImageNet labels, they can also potentially benefit from self-supervised learning techniques such as masked autoencoders (MAE). However, we found that simply combining these two approaches leads to subpar performance. In this paper, we propose a fully convolutional masked autoencoder framework and a new Global Response Normalization (GRN) layer that can be added to the ConvNeXt architecture to enhance inter-channel feature competition. This co-design of self-supervised learning techniques and architectural improvement results in a new model family called ConvNeXt V2, which significantly improves the performance of pure ConvNets on various recognition benchmarks, including ImageNet classification, COCO detection, and ADE20K segmentation. We also provide pre-trained ConvNeXt V2 models of various sizes, ranging from an efficient 3.7M-parameter Atto model with 76.7% top-1 accuracy on ImageNet, to a 650M Huge model that achieves a state-of-the-art 88.9% accuracy using only public training data.

Sanghyun Woo, Kwanyong Park, Seoung Wug Oh et al.

The tracking-by-detection paradigm today has become the dominant method for multi-object tracking and works by detecting objects in each frame and then performing data association across frames. However, its sequential frame-wise matching property fundamentally suffers from the intermediate interruptions in a video, such as object occlusions, fast camera movements, and abrupt light changes. Moreover, it typically overlooks temporal information beyond the two frames for matching. In this paper, we investigate an alternative by treating object association as clip-wise matching. Our new perspective views a single long video sequence as multiple short clips, and then the tracking is performed both within and between the clips. The benefits of this new approach are two folds. First, our method is robust to tracking error accumulation or propagation, as the video chunking allows bypassing the interrupted frames, and the short clip tracking avoids the conventional error-prone long-term track memory management. Second, the multiple frame information is aggregated during the clip-wise matching, resulting in a more accurate long-range track association than the current frame-wise matching. Given the state-of-the-art tracking-by-detection tracker, QDTrack, we showcase how the tracking performance improves with our new tracking formulation. We evaluate our proposals on two tracking benchmarks, TAO and MOT17 that have complementary characteristics and challenges each other.

Kwanyong Park, Sanghyun Woo, Seoung Wug Oh et al.

Recently, memory-based approaches show promising results on semi-supervised video object segmentation. These methods predict object masks frame-by-frame with the help of frequently updated memory of the previous mask. Different from this per-frame inference, we investigate an alternative perspective by treating video object segmentation as clip-wise mask propagation. In this per-clip inference scheme, we update the memory with an interval and simultaneously process a set of consecutive frames (i.e. clip) between the memory updates. The scheme provides two potential benefits: accuracy gain by clip-level optimization and efficiency gain by parallel computation of multiple frames. To this end, we propose a new method tailored for the per-clip inference. Specifically, we first introduce a clip-wise operation to refine the features based on intra-clip correlation. In addition, we employ a progressive matching mechanism for efficient information-passing within a clip. With the synergy of two modules and a newly proposed per-clip based training, our network achieves state-of-the-art performance on Youtube-VOS 2018/2019 val (84.6% and 84.6%) and DAVIS 2016/2017 val (91.9% and 86.1%). Furthermore, our model shows a great speed-accuracy trade-off with varying memory update intervals, which leads to huge flexibility.

Sanghyun Woo, Kwanyong Park, Seoung Wug Oh et al.

Scaling object taxonomies is one of the important steps toward a robust real-world deployment of recognition systems. We have faced remarkable progress in images since the introduction of the LVIS benchmark. To continue this success in videos, a new video benchmark, TAO, was recently presented. Given the recent encouraging results from both detection and tracking communities, we are interested in marrying those two advances and building a strong large vocabulary video tracker. However, supervisions in LVIS and TAO are inherently sparse or even missing, posing two new challenges for training the large vocabulary trackers. First, no tracking supervisions are in LVIS, which leads to inconsistent learning of detection (with LVIS and TAO) and tracking (only with TAO). Second, the detection supervisions in TAO are partial, which results in catastrophic forgetting of absent LVIS categories during video fine-tuning. To resolve these challenges, we present a simple but effective learning framework that takes full advantage of all available training data to learn detection and tracking while not losing any LVIS categories to recognize. With this new learning scheme, we show that consistent improvements of various large vocabulary trackers are capable, setting strong baseline results on the challenging TAO benchmarks.

Fu-Yun Wang, Hao Zhou, Liangzhe Yuan et al.

The slow inference process of image diffusion models significantly degrades interactive user experiences. To address this, we introduce Diffusion Preview, a novel paradigm employing rapid, low-step sampling to generate preliminary outputs for user evaluation, deferring full-step refinement until the preview is deemed satisfactory. Existing acceleration methods, including training-free solvers and post-training distillation, struggle to deliver high-quality previews or ensure consistency between previews and final outputs. We propose ConsistencySolver derived from general linear multistep methods, a lightweight, trainable high-order solver optimized via Reinforcement Learning, that enhances preview quality and consistency. Experimental results demonstrate that ConsistencySolver significantly improves generation quality and consistency in low-step scenarios, making it ideal for efficient preview-and-refine workflows. Notably, it achieves FID scores on-par with Multistep DPM-Solver using 47% fewer steps, while outperforming distillation baselines. Furthermore, user studies indicate our approach reduces overall user interaction time by nearly 50% while maintaining generation quality. Code is available at https://github.com/G-U-N/consolver.

Daehan Kim, Minseok Seo, Kwanyong Park et al.

Mixup provides interpolated training samples and allows the model to obtain smoother decision boundaries for better generalization. The idea can be naturally applied to the domain adaptation task, where we can mix the source and target samples to obtain domain-mixed samples for better adaptation. However, the extension of the idea from classification to segmentation (i.e., structured output) is nontrivial. This paper systematically studies the impact of mixup under the domain adaptaive semantic segmentation task and presents a simple yet effective mixup strategy called Bidirectional Domain Mixup (BDM). In specific, we achieve domain mixup in two-step: cut and paste. Given the warm-up model trained from any adaptation techniques, we forward the source and target samples and perform a simple threshold-based cut out of the unconfident regions (cut). After then, we fill-in the dropped regions with the other domain region patches (paste). In doing so, we jointly consider class distribution, spatial structure, and pseudo label confidence. Based on our analysis, we found that BDM leaves domain transferable regions by cutting, balances the dataset-level class distribution while preserving natural scene context by pasting. We coupled our proposal with various state-of-the-art adaptation models and observe significant improvement consistently. We also provide extensive ablation experiments to empirically verify our main components of the framework. Visit our project page with the code at https://sites.google.com/view/bidirectional-domain-mixup

Sungsu Hur, Inkyu Shin, Kwanyong Park et al.

Universal Domain Adaptation aims to transfer the knowledge between the datasets by handling two shifts: domain-shift and category-shift. The main challenge is correctly distinguishing the unknown target samples while adapting the distribution of known class knowledge from source to target. Most existing methods approach this problem by first training the target adapted known classifier and then relying on the single threshold to distinguish unknown target samples. However, this simple threshold-based approach prevents the model from considering the underlying complexities existing between the known and unknown samples in the high-dimensional feature space. In this paper, we propose a new approach in which we use two sets of feature points, namely dual Classifiers for Prototypes and Reciprocals (CPR). Our key idea is to associate each prototype with corresponding known class features while pushing the reciprocals apart from these prototypes to locate them in the potential unknown feature space. The target samples are then classified as unknown if they fall near any reciprocals at test time. To successfully train our framework, we collect the partial, confident target samples that are classified as known or unknown through on our proposed multi-criteria selection. We then additionally apply the entropy loss regularization to them. For further adaptation, we also apply standard consistency regularization that matches the predictions of two different views of the input to make more compact target feature space. We evaluate our proposal, CPR, on three standard benchmarks and achieve comparable or new state-of-the-art results. We also provide extensive ablation experiments to verify our main design choices in our framework.

Junha Song, Kwanyong Park, InKyu Shin et al.

Prior to the deployment of robotic systems, pre-training the deep-recognition models on all potential visual cases is infeasible in practice. Hence, test-time adaptation (TTA) allows the model to adapt itself to novel environments and improve its performance during test time (i.e., lifelong adaptation). Several works for TTA have shown promising adaptation performances in continuously changing environments. However, our investigation reveals that existing methods are vulnerable to dynamic distributional changes and often lead to overfitting of TTA models. To address this problem, this paper first presents a robust TTA framework with compound domain knowledge management. Our framework helps the TTA model to harvest the knowledge of multiple representative domains (i.e., compound domain) and conduct the TTA based on the compound domain knowledge. In addition, to prevent overfitting of the TTA model, we devise novel regularization which modulates the adaptation rates using domain-similarity between the source and the current target domain. With the synergy of the proposed framework and regularization, we achieve consistent performance improvements in diverse TTA scenarios, especially on dynamic domain shifts. We demonstrate the generality of proposals via extensive experiments including image classification on ImageNet-C and semantic segmentation on GTA5, C-driving, and corrupted Cityscapes datasets.

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Zhengfei Kuang, Rui Lin, Long Zhao et al.

Despite the remarkable progress of Multimodal Large Language Models (MLLMs) in 2D vision-language tasks, their application to complex 3D scene manipulation remains underexplored. In this paper, we bridge this critical gap by tackling three key challenges in 3D object arrangement task using MLLMs. First, to address the weak visual grounding of MLLMs, which struggle to link programmatic edits with precise 3D outcomes, we introduce an MCP-based API. This shifts the interaction from brittle raw code manipulation to more robust, function-level updates. Second, we augment the MLLM's 3D scene understanding with a suite of specialized visual tools to analyze scene state, gather spatial information, and validate action outcomes. This perceptual feedback loop is critical for closing the gap between language-based updates and precise 3D-aware manipulation. Third, to manage the iterative, error-prone updates, we propose a collaborative multi-agent framework with designated roles for planning, execution, and verification. This decomposition allows the system to robustly handle multi-step instructions and recover from intermediate errors. We demonstrate the effectiveness of our approach on a diverse set of 25 complex object arrangement tasks, where it significantly outperforms existing baselines. Website: vulcan-3d.github.io

Hoon Kim, Minje Jang, Wonjun Yoon et al.

We introduce a co-designed approach for human portrait relighting that combines a physics-guided architecture with a pre-training framework. Drawing on the Cook-Torrance reflectance model, we have meticulously configured the architecture design to precisely simulate light-surface interactions. Furthermore, to overcome the limitation of scarce high-quality lightstage data, we have developed a self-supervised pre-training strategy. This novel combination of accurate physical modeling and expanded training dataset establishes a new benchmark in relighting realism.

Jiacheng Chen, Ramin Mehran, Xuhui Jia et al.

We present BlenderFusion, a generative visual compositing framework that synthesizes new scenes by recomposing objects, camera, and background. It follows a layering-editing-compositing pipeline: (i) segmenting and converting visual inputs into editable 3D entities (layering), (ii) editing them in Blender with 3D-grounded control (editing), and (iii) fusing them into a coherent scene using a generative compositor (compositing). Our generative compositor extends a pre-trained diffusion model to process both the original (source) and edited (target) scenes in parallel. It is fine-tuned on video frames with two key training strategies: (i) source masking, enabling flexible modifications like background replacement; (ii) simulated object jittering, facilitating disentangled control over objects and camera. BlenderFusion significantly outperforms prior methods in complex compositional scene editing tasks.

Sanghyun Woo, Kwanyong Park, Inkyu Shin et al.

Multi-target multi-camera tracking is a crucial task that involves identifying and tracking individuals over time using video streams from multiple cameras. This task has practical applications in various fields, such as visual surveillance, crowd behavior analysis, and anomaly detection. However, due to the difficulty and cost of collecting and labeling data, existing datasets for this task are either synthetically generated or artificially constructed within a controlled camera network setting, which limits their ability to model real-world dynamics and generalize to diverse camera configurations. To address this issue, we present MTMMC, a real-world, large-scale dataset that includes long video sequences captured by 16 multi-modal cameras in two different environments - campus and factory - across various time, weather, and season conditions. This dataset provides a challenging test-bed for studying multi-camera tracking under diverse real-world complexities and includes an additional input modality of spatially aligned and temporally synchronized RGB and thermal cameras, which enhances the accuracy of multi-camera tracking. MTMMC is a super-set of existing datasets, benefiting independent fields such as person detection, re-identification, and multiple object tracking. We provide baselines and new learning setups on this dataset and set the reference scores for future studies. The datasets, models, and test server will be made publicly available.

Shengbang Tong, Ellis Brown, Penghao Wu et al.

We introduce Cambrian-1, a family of multimodal LLMs (MLLMs) designed with a vision-centric approach. While stronger language models can enhance multimodal capabilities, the design choices for vision components are often insufficiently explored and disconnected from visual representation learning research. This gap hinders accurate sensory grounding in real-world scenarios. Our study uses LLMs and visual instruction tuning as an interface to evaluate various visual representations, offering new insights into different models and architectures -- self-supervised, strongly supervised, or combinations thereof -- based on experiments with over 20 vision encoders. We critically examine existing MLLM benchmarks, address the difficulties involved in consolidating and interpreting results from various tasks, and introduce a new vision-centric benchmark, CV-Bench. To further improve visual grounding, we propose the Spatial Vision Aggregator (SVA), a dynamic and spatially-aware connector that integrates high-resolution vision features with LLMs while reducing the number of tokens. Additionally, we discuss the curation of high-quality visual instruction-tuning data from publicly available sources, emphasizing the importance of data source balancing and distribution ratio. Collectively, Cambrian-1 not only achieves state-of-the-art performance but also serves as a comprehensive, open cookbook for instruction-tuned MLLMs. We provide model weights, code, supporting tools, datasets, and detailed instruction-tuning and evaluation recipes. We hope our release will inspire and accelerate advancements in multimodal systems and visual representation learning.

KwanYong Park, Sanghyun Woo, Inkyu Shin et al.

Unsupervised domain adaptation (UDA) for semantic segmentation has been attracting attention recently, as it could be beneficial for various label-scarce real-world scenarios (e.g., robot control, autonomous driving, medical imaging, etc.). Despite the significant progress in this field, current works mainly focus on a single-source single-target setting, which cannot handle more practical settings of multiple targets or even unseen targets. In this paper, we investigate open compound domain adaptation (OCDA), which deals with mixed and novel situations at the same time, for semantic segmentation. We present a novel framework based on three main design principles: discover, hallucinate, and adapt. The scheme first clusters compound target data based on style, discovering multiple latent domains (discover). Then, it hallucinates multiple latent target domains in source by using image-translation (hallucinate). This step ensures the latent domains in the source and the target to be paired. Finally, target-to-source alignment is learned separately between domains (adapt). In high-level, our solution replaces a hard OCDA problem with much easier multiple UDA problems. We evaluate our solution on standard benchmark GTA to C-driving, and achieved new state-of-the-art results.

Adrit Rao, Jongchan Park, Sanghyun Woo et al.

When the trained physician interprets medical images, they understand the clinical importance of visual features. By applying cognitive attention, they apply greater focus onto clinically relevant regions while disregarding unnecessary features. The use of computer vision to automate the classification of medical images is widely studied. However, the standard convolutional neural network (CNN) does not necessarily employ subconscious feature relevancy evaluation techniques similar to the trained medical specialist and evaluates features more generally. Self-attention mechanisms enable CNNs to focus more on semantically important regions or aggregated relevant context with long-range dependencies. By using attention, medical image analysis systems can potentially become more robust by focusing on more important clinical feature regions. In this paper, we provide a comprehensive comparison of various state-of-the-art self-attention mechanisms across multiple medical image analysis tasks. Through both quantitative and qualitative evaluations along with a clinical user-centric survey study, we aim to provide a deeper understanding of the effects of self-attention in medical computer vision tasks.

Inkyu Shin, Dong-jin Kim, Jae Won Cho et al.

Unsupervised Domain Adaptation (UDA) for semantic segmentation has been actively studied to mitigate the domain gap between label-rich source data and unlabeled target data. Despite these efforts, UDA still has a long way to go to reach the fully supervised performance. To this end, we propose a Labeling Only if Required strategy, LabOR, where we introduce a human-in-the-loop approach to adaptively give scarce labels to points that a UDA model is uncertain about. In order to find the uncertain points, we generate an inconsistency mask using the proposed adaptive pixel selector and we label these segment-based regions to achieve near supervised performance with only a small fraction (about 2.2%) ground truth points, which we call "Segment based Pixel-Labeling (SPL)". To further reduce the efforts of the human annotator, we also propose "Point-based Pixel-Labeling (PPL)", which finds the most representative points for labeling within the generated inconsistency mask. This reduces efforts from 2.2% segment label to 40 points label while minimizing performance degradation. Through extensive experimentation, we show the advantages of this new framework for domain adaptive semantic segmentation while minimizing human labor costs.

Inkyu Shin, Kwanyong Park, Sanghyun Woo et al.

Unsupervised Domain Adaptation for semantic segmentation has gained immense popularity since it can transfer knowledge from simulation to real (Sim2Real) by largely cutting out the laborious per pixel labeling efforts at real. In this work, we present a new video extension of this task, namely Unsupervised Domain Adaptation for Video Semantic Segmentation. As it became easy to obtain large-scale video labels through simulation, we believe attempting to maximize Sim2Real knowledge transferability is one of the promising directions for resolving the fundamental data-hungry issue in the video. To tackle this new problem, we present a novel two-phase adaptation scheme. In the first step, we exhaustively distill source domain knowledge using supervised loss functions. Simultaneously, video adversarial training (VAT) is employed to align the features from source to target utilizing video context. In the second step, we apply video self-training (VST), focusing only on the target data. To construct robust pseudo labels, we exploit the temporal information in the video, which has been rarely explored in the previous image-based self-training approaches. We set strong baseline scores on 'VIPER to CityscapeVPS' adaptation scenario. We show that our proposals significantly outperform previous image-based UDA methods both on image-level (mIoU) and video-level (VPQ) evaluation metrics.

Sanghyun Woo, Dahun Kim, Joon-Young Lee et al.

Temporal correspondence - linking pixels or objects across frames - is a fundamental supervisory signal for the video models. For the panoptic understanding of dynamic scenes, we further extend this concept to every segment. Specifically, we aim to learn coarse segment-level matching and fine pixel-level matching together. We implement this idea by designing two novel learning objectives. To validate our proposals, we adopt a deep siamese model and train the model to learn the temporal correspondence on two different levels (i.e., segment and pixel) along with the target task. At inference time, the model processes each frame independently without any extra computation and post-processing. We show that our per-frame inference model can achieve new state-of-the-art results on Cityscapes-VPS and VIPER datasets. Moreover, due to its high efficiency, the model runs in a fraction of time (3x) compared to the previous state-of-the-art approach.

Inkyu Shin, Sanghyun Woo, Fei Pan et al.

Recently, deep self-training approaches emerged as a powerful solution to the unsupervised domain adaptation. The self-training scheme involves iterative processing of target data; it generates target pseudo labels and retrains the network. However, since only the confident predictions are taken as pseudo labels, existing self-training approaches inevitably produce sparse pseudo labels in practice. We see this is critical because the resulting insufficient training-signals lead to a suboptimal, error-prone model. In order to tackle this problem, we propose a novel Two-phase Pseudo Label Densification framework, referred to as TPLD. In the first phase, we use sliding window voting to propagate the confident predictions, utilizing intrinsic spatial-correlations in the images. In the second phase, we perform a confidence-based easy-hard classification. For the easy samples, we now employ their full pseudo labels. For the hard ones, we instead adopt adversarial learning to enforce hard-to-easy feature alignment. To ease the training process and avoid noisy predictions, we introduce the bootstrapping mechanism to the original self-training loss. We show the proposed TPLD can be easily integrated into existing self-training based approaches and improves the performance significantly. Combined with the recently proposed CRST self-training framework, we achieve new state-of-the-art results on two standard UDA benchmarks.

Myungchul Kim, Sanghyun Woo, Dahun Kim et al.

Pursuing a more coherent scene understanding towards real-time vision applications, single-stage instance segmentation has recently gained popularity, achieving a simpler and more efficient design than its two-stage counterparts. Besides, its global mask representation often leads to superior accuracy to the two-stage Mask R-CNN which has been dominant thus far. Despite the promising advances in single-stage methods, finer delineation of instance boundaries still remains unexcavated. Indeed, boundary information provides a strong shape representation that can operate in synergy with the fully-convolutional mask features of the single-stage segmenter. In this work, we propose Boundary Basis based Instance Segmentation(B2Inst) to learn a global boundary representation that can complement existing global-mask-based methods that are often lacking high-frequency details. Besides, we devise a unified quality measure of both mask and boundary and introduce a network block that learns to score the per-instance predictions of itself. When applied to the strongest baselines in single-stage instance segmentation, our B2Inst leads to consistent improvements and accurately parse out the instance boundaries in a scene. Regardless of being single-stage or two-stage frameworks, we outperform the existing state-of-the-art methods on the COCO dataset with the same ResNet-50 and ResNet-101 backbones.

Dahun Kim, Sanghyun Woo, Joon-Young Lee et al.

Panoptic segmentation has become a new standard of visual recognition task by unifying previous semantic segmentation and instance segmentation tasks in concert. In this paper, we propose and explore a new video extension of this task, called video panoptic segmentation. The task requires generating consistent panoptic segmentation as well as an association of instance ids across video frames. To invigorate research on this new task, we present two types of video panoptic datasets. The first is a re-organization of the synthetic VIPER dataset into the video panoptic format to exploit its large-scale pixel annotations. The second is a temporal extension on the Cityscapes val. set, by providing new video panoptic annotations (Cityscapes-VPS). Moreover, we propose a novel video panoptic segmentation network (VPSNet) which jointly predicts object classes, bounding boxes, masks, instance id tracking, and semantic segmentation in video frames. To provide appropriate metrics for this task, we propose a video panoptic quality (VPQ) metric and evaluate our method and several other baselines. Experimental results demonstrate the effectiveness of the presented two datasets. We achieve state-of-the-art results in image PQ on Cityscapes and also in VPQ on Cityscapes-VPS and VIPER datasets. The datasets and code are made publicly available.

Yunjae Jung, Dahun Kim, Sanghyun Woo et al.

Visual storytelling is a task of creating a short story based on photo streams. Unlike existing visual captioning, storytelling aims to contain not only factual descriptions, but also human-like narration and semantics. However, the VIST dataset consists only of a small, fixed number of photos per story. Therefore, the main challenge of visual storytelling is to fill in the visual gap between photos with narrative and imaginative story. In this paper, we propose to explicitly learn to imagine a storyline that bridges the visual gap. During training, one or more photos is randomly omitted from the input stack, and we train the network to produce a full plausible story even with missing photo(s). Furthermore, we propose for visual storytelling a hide-and-tell model, which is designed to learn non-local relations across the photo streams and to refine and improve conventional RNN-based models. In experiments, we show that our scheme of hide-and-tell, and the network design are indeed effective at storytelling, and that our model outperforms previous state-of-the-art methods in automatic metrics. Finally, we qualitatively show the learned ability to interpolate storyline over visual gaps.

Kwanyong Park, Sanghyun Woo, Dahun Kim et al.

In this paper, we investigate the problem of unpaired video-to-video translation. Given a video in the source domain, we aim to learn the conditional distribution of the corresponding video in the target domain, without seeing any pairs of corresponding videos. While significant progress has been made in the unpaired translation of images, directly applying these methods to an input video leads to low visual quality due to the additional time dimension. In particular, previous methods suffer from semantic inconsistency (i.e., semantic label flipping) and temporal flickering artifacts. To alleviate these issues, we propose a new framework that is composed of carefully-designed generators and discriminators, coupled with two core objective functions: 1) content preserving loss and 2) temporal consistency loss. Extensive qualitative and quantitative evaluations demonstrate the superior performance of the proposed method against previous approaches. We further apply our framework to a domain adaptation task and achieve favorable results.

Ho-Deok Jang, Sanghyun Woo, Philipp Benz et al.

We present a simple yet effective prediction module for a one-stage detector. The main process is conducted in a coarse-to-fine manner. First, the module roughly adjusts the default boxes to well capture the extent of target objects in an image. Second, given the adjusted boxes, the module aligns the receptive field of the convolution filters accordingly, not requiring any embedding layers. Both steps build a propose-and-attend mechanism, mimicking two-stage detectors in a highly efficient manner. To verify its effectiveness, we apply the proposed module to a basic one-stage detector SSD. Our final model achieves an accuracy comparable to that of state-of-the-art detectors while using a fraction of their model parameters and computational overheads. Moreover, we found that the proposed module has two strong applications. 1) The module can be successfully integrated into a lightweight backbone, further pushing the efficiency of the one-stage detector. 2) The module also allows train-from-scratch without relying on any sophisticated base networks as previous methods do.

Sanghyun Woo, Dahun Kim, KwanYong Park et al.

We propose a novel feed-forward network for video inpainting. We use a set of sampled video frames as the reference to take visible contents to fill the hole of a target frame. Our video inpainting network consists of two stages. The first stage is an alignment module that uses computed homographies between the reference frames and the target frame. The visible patches are then aggregated based on the frame similarity to fill in the target holes roughly. The second stage is a non-local attention module that matches the generated patches with known reference patches (in space and time) to refine the previous global alignment stage. Both stages consist of large spatial-temporal window size for the reference and thus enable modeling long-range correlations between distant information and the hole regions. Therefore, even challenging scenes with large or slowly moving holes can be handled, which have been hardly modeled by existing flow-based approach. Our network is also designed with a recurrent propagation stream to encourage temporal consistency in video results. Experiments on video object removal demonstrate that our method inpaints the holes with globally and locally coherent contents.

Dahun Kim, Sanghyun Woo, Joon-Young Lee et al.

Blind video decaptioning is a problem of automatically removing text overlays and inpainting the occluded parts in videos without any input masks. While recent deep learning based inpainting methods deal with a single image and mostly assume that the positions of the corrupted pixels are known, we aim at automatic text removal in video sequences without mask information. In this paper, we propose a simple yet effective framework for fast blind video decaptioning. We construct an encoder-decoder model, where the encoder takes multiple source frames that can provide visible pixels revealed from the scene dynamics. These hints are aggregated and fed into the decoder. We apply a residual connection from the input frame to the decoder output to enforce our network to focus on the corrupted regions only. Our proposed model was ranked in the first place in the ECCV Chalearn 2018 LAP Inpainting Competition Track2: Video decaptioning. In addition, we further improve this strong model by applying a recurrent feedback. The recurrent feedback not only enforces temporal coherence but also provides strong clues on where the corrupted pixels are. Both qualitative and quantitative experiments demonstrate that our full model produces accurate and temporally consistent video results in real time (50+ fps).

Dahun Kim, Sanghyun Woo, Joon-Young Lee et al.

Video inpainting aims to fill spatio-temporal holes with plausible content in a video. Despite tremendous progress of deep neural networks for image inpainting, it is challenging to extend these methods to the video domain due to the additional time dimension. In this work, we propose a novel deep network architecture for fast video inpainting. Built upon an image-based encoder-decoder model, our framework is designed to collect and refine information from neighbor frames and synthesize still-unknown regions. At the same time, the output is enforced to be temporally consistent by a recurrent feedback and a temporal memory module. Compared with the state-of-the-art image inpainting algorithm, our method produces videos that are much more semantically correct and temporally smooth. In contrast to the prior video completion method which relies on time-consuming optimization, our method runs in near real-time while generating competitive video results. Finally, we applied our framework to video retargeting task, and obtain visually pleasing results.

Yunjae Jung, Donghyeon Cho, Dahun Kim et al.

In this paper, we address the problem of unsupervised video summarization that automatically extracts key-shots from an input video. Specifically, we tackle two critical issues based on our empirical observations: (i) Ineffective feature learning due to flat distributions of output importance scores for each frame, and (ii) training difficulty when dealing with long-length video inputs. To alleviate the first problem, we propose a simple yet effective regularization loss term called variance loss. The proposed variance loss allows a network to predict output scores for each frame with high discrepancy which enables effective feature learning and significantly improves model performance. For the second problem, we design a novel two-stream network named Chunk and Stride Network (CSNet) that utilizes local (chunk) and global (stride) temporal view on the video features. Our CSNet gives better summarization results for long-length videos compared to the existing methods. In addition, we introduce an attention mechanism to handle the dynamic information in videos. We demonstrate the effectiveness of the proposed methods by conducting extensive ablation studies and show that our final model achieves new state-of-the-art results on two benchmark datasets.

Sanghyun Woo, Dahun Kim, Donghyeon Cho et al.

Objects and their relationships are critical contents for image understanding. A scene graph provides a structured description that captures these properties of an image. However, reasoning about the relationships between objects is very challenging and only a few recent works have attempted to solve the problem of generating a scene graph from an image. In this paper, we present a method that improves scene graph generation by explicitly modeling inter-dependency among the entire object instances. We design a simple and effective relational embedding module that enables our model to jointly represent connections among all related objects, rather than focus on an object in isolation. Our method significantly benefits the main part of the scene graph generation task: relationship classification. Using it on top of a basic Faster R-CNN, our model achieves state-of-the-art results on the Visual Genome benchmark. We further push the performance by introducing global context encoding module and geometrical layout encoding module. We validate our final model, LinkNet, through extensive ablation studies, demonstrating its efficacy in scene graph generation.

Sanghyun Woo, Jongchan Park, Joon-Young Lee et al.

We propose Convolutional Block Attention Module (CBAM), a simple yet effective attention module for feed-forward convolutional neural networks. Given an intermediate feature map, our module sequentially infers attention maps along two separate dimensions, channel and spatial, then the attention maps are multiplied to the input feature map for adaptive feature refinement. Because CBAM is a lightweight and general module, it can be integrated into any CNN architectures seamlessly with negligible overheads and is end-to-end trainable along with base CNNs. We validate our CBAM through extensive experiments on ImageNet-1K, MS~COCO detection, and VOC~2007 detection datasets. Our experiments show consistent improvements in classification and detection performances with various models, demonstrating the wide applicability of CBAM. The code and models will be publicly available.

Jongchan Park, Sanghyun Woo, Joon-Young Lee et al.

Recent advances in deep neural networks have been developed via architecture search for stronger representational power. In this work, we focus on the effect of attention in general deep neural networks. We propose a simple and effective attention module, named Bottleneck Attention Module (BAM), that can be integrated with any feed-forward convolutional neural networks. Our module infers an attention map along two separate pathways, channel and spatial. We place our module at each bottleneck of models where the downsampling of feature maps occurs. Our module constructs a hierarchical attention at bottlenecks with a number of parameters and it is trainable in an end-to-end manner jointly with any feed-forward models. We validate our BAM through extensive experiments on CIFAR-100, ImageNet-1K, VOC 2007 and MS COCO benchmarks. Our experiments show consistent improvement in classification and detection performances with various models, demonstrating the wide applicability of BAM. The code and models will be publicly available.

Sanghyun Woo, Soonmin Hwang, In So Kweon

One-stage object detectors such as SSD or YOLO already have shown promising accuracy with small memory footprint and fast speed. However, it is widely recognized that one-stage detectors have difficulty in detecting small objects while they are competitive with two-stage methods on large objects. In this paper, we investigate how to alleviate this problem starting from the SSD framework. Due to their pyramidal design, the lower layer that is responsible for small objects lacks strong semantics(e.g contextual information). We address this problem by introducing a feature combining module that spreads out the strong semantics in a top-down manner. Our final model StairNet detector unifies the multi-scale representations and semantic distribution effectively. Experiments on PASCAL VOC 2007 and PASCAL VOC 2012 datasets demonstrate that StairNet significantly improves the weakness of SSD and outperforms the other state-of-the-art one-stage detectors.