Minseok Seo

Dina Bashkirova, Samarth Mishra, Diala Lteif et al.

Label-efficient and reliable semantic segmentation is essential for many real-life applications, especially for industrial settings with high visual diversity, such as waste sorting. In industrial waste sorting, one of the biggest challenges is the extreme diversity of the input stream depending on factors like the location of the sorting facility, the equipment available in the facility, and the time of year, all of which significantly impact the composition and visual appearance of the waste stream. These changes in the data are called ``visual domains'', and label-efficient adaptation of models to such domains is needed for successful semantic segmentation of industrial waste. To test the abilities of computer vision models on this task, we present the VisDA 2022 Challenge on Domain Adaptation for Industrial Waste Sorting. Our challenge incorporates a fully-annotated waste sorting dataset, ZeroWaste, collected from two real material recovery facilities in different locations and seasons, as well as a novel procedurally generated synthetic waste sorting dataset, SynthWaste. In this competition, we aim to answer two questions: 1) can we leverage domain adaptation techniques to minimize the domain gap? and 2) can synthetic data augmentation improve performance on this task and help adapt to changing data distributions? The results of the competition show that industrial waste detection poses a real domain adaptation problem, that domain generalization techniques such as augmentations, ensembling, etc., improve the overall performance on the unlabeled target domain examples, and that leveraging synthetic data effectively remains an open problem. See https://ai.bu.edu/visda-2022/

Daehan Kim, Minseok Seo, YoungJin Jeon et al.

The Visual Domain Adaptation(VisDA) 2022 Challenge calls for an unsupervised domain adaptive model in semantic segmentation tasks for industrial waste sorting. In this paper, we introduce the SIA_Adapt method, which incorporates several methods for domain adaptive models. The core of our method in the transferable representation from large-scale pre-training. In this process, we choose a network architecture that differs from the state-of-the-art for domain adaptation. After that, self-training using pseudo-labels helps to make the initial adaptation model more adaptable to the target domain. Finally, the model soup scheme helped to improve the generalization performance in the target domain. Our method SIA_Adapt achieves 1st place in the VisDA2022 challenge. The code is available on https: //github.com/DaehanKim-Korea/VisDA2022_Winner_Solution.

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

Hyeoncheol Noh, Jingi Ju, Minseok Seo et al.

To train the change detector, bi-temporal images taken at different times in the same area are used. However, collecting labeled bi-temporal images is expensive and time consuming. To solve this problem, various unsupervised change detection methods have been proposed, but they still require unlabeled bi-temporal images. In this paper, we propose unsupervised change detection based on image reconstruction loss using only unlabeled single temporal single image. The image reconstruction model is trained to reconstruct the original source image by receiving the source image and the photometrically transformed source image as a pair. During inference, the model receives bi-temporal images as the input, and tries to reconstruct one of the inputs. The changed region between bi-temporal images shows high reconstruction loss. Our change detector showed significant performance in various change detection benchmark datasets even though only a single temporal single source image was used. The code and trained models will be publicly available for reproducibility.

Minseok Seo, Hakjin Lee, Doyi Kim et al.

Future frame prediction has been approached through two primary methods: autoregressive and non-autoregressive. Autoregressive methods rely on the Markov assumption and can achieve high accuracy in the early stages of prediction when errors are not yet accumulated. However, their performance tends to decline as the number of time steps increases. In contrast, non-autoregressive methods can achieve relatively high performance but lack correlation between predictions for each time step. In this paper, we propose an Implicit Stacked Autoregressive Model for Video Prediction (IAM4VP), which is an implicit video prediction model that applies a stacked autoregressive method. Like non-autoregressive methods, stacked autoregressive methods use the same observed frame to estimate all future frames. However, they use their own predictions as input, similar to autoregressive methods. As the number of time steps increases, predictions are sequentially stacked in the queue. To evaluate the effectiveness of IAM4VP, we conducted experiments on three common future frame prediction benchmark datasets and weather\&climate prediction benchmark datasets. The results demonstrate that our proposed model achieves state-of-the-art performance.

Daehan Kim, Minseok Seo, Jinsun Park et al.

In this paper, we introduce source domain subset sampling (SDSS) as a new perspective of semi-supervised domain adaptation. We propose domain adaptation by sampling and exploiting only a meaningful subset from source data for training. Our key assumption is that the entire source domain data may contain samples that are unhelpful for the adaptation. Therefore, the domain adaptation can benefit from a subset of source data composed solely of helpful and relevant samples. The proposed method effectively subsamples full source data to generate a small-scale meaningful subset. Therefore, training time is reduced, and performance is improved with our subsampled source data. To further verify the scalability of our method, we construct a new dataset called Ocean Ship, which comprises 500 real and 200K synthetic sample images with ground-truth labels. The SDSS achieved a state-of-the-art performance when applied on GTA5 to Cityscapes and SYNTHIA to Cityscapes public benchmark datasets and a 9.13 mIoU improvement on our Ocean Ship dataset over a baseline model.

Minseok Seo, Yeji Choi, Hyungon Ry et al.

Geostationary satellite imagery has applications in climate and weather forecasting, planning natural energy resources, and predicting extreme weather events. For precise and accurate prediction, higher spatial and temporal resolution of geostationary satellite imagery is important. Although recent geostationary satellite resolution has improved, the long-term analysis of climate applications is limited to using multiple satellites from the past to the present due to the different resolutions. To solve this problem, we proposed warp and refine network (WR-Net). WR-Net is divided into an optical flow warp component and a warp image refinement component. We used the TV-L1 algorithm instead of deep learning-based approaches to extract the optical flow warp component. The deep-learning-based model is trained on the human-centric view of the RGB channel and does not work on geostationary satellites, which is gray-scale one-channel imagery. The refinement network refines the warped image through a multi-temporal fusion layer. We evaluated WR-Net by interpolation of temporal resolution at 4 min intervals to 2 min intervals in large-scale GK2A geostationary meteorological satellite imagery. Furthermore, we applied WR-Net to the future frame prediction task and showed that the explicit use of optical flow can help future frame prediction.

Donggeun Yoon, Minseok Seo, Doyi Kim et al.

Weather forecasting requires not only accuracy but also the ability to perform probabilistic prediction. However, deterministic weather forecasting methods do not support probabilistic predictions, and conversely, probabilistic models tend to be less accurate. To address these challenges, in this paper, we introduce the \textbf{\textit{D}}eterministic \textbf{\textit{G}}uidance \textbf{\textit{D}}iffusion \textbf{\textit{M}}odel (DGDM) for probabilistic weather forecasting, integrating benefits of both deterministic and probabilistic approaches. During the forward process, both the deterministic and probabilistic models are trained end-to-end. In the reverse process, weather forecasting leverages the predicted result from the deterministic model, using as an intermediate starting point for the probabilistic model. By fusing deterministic models with probabilistic models in this manner, DGDM is capable of providing accurate forecasts while also offering probabilistic predictions. To evaluate DGDM, we assess it on the global weather forecasting dataset (WeatherBench) and the common video frame prediction benchmark (Moving MNIST). We also introduce and evaluate the Pacific Northwest Windstorm (PNW)-Typhoon weather satellite dataset to verify the effectiveness of DGDM in high-resolution regional forecasting. As a result of our experiments, DGDM achieves state-of-the-art results not only in global forecasting but also in regional forecasting. The code is available at: \url{https://github.com/DongGeun-Yoon/DGDM}.

Jaemin Lee, Minseok Seo, Sangwoo Lee et al.

In general, deep learning-based video frame interpolation (VFI) methods have predominantly focused on estimating motion vectors between two input frames and warping them to the target time. While this approach has shown impressive performance for linear motion between two input frames, it exhibits limitations when dealing with occlusions and nonlinear movements. Recently, generative models have been applied to VFI to address these issues. However, as VFI is not a task focused on generating plausible images, but rather on predicting accurate intermediate frames between two given frames, performance limitations still persist. In this paper, we propose a multi-in-single-out (MISO) based VFI method that does not rely on motion vector estimation, allowing it to effectively model occlusions and nonlinear motion. Additionally, we introduce a novel motion perceptual loss that enables MISO-VFI to better capture the spatio-temporal correlations within the video frames. Our MISO-VFI method achieves state-of-the-art results on VFI benchmarks Vimeo90K, Middlebury, and UCF101, with a significant performance gap compared to existing approaches.

Minseok Seo, Wonjun Lee, Jaehyuk Jang et al.

Zero-shot depth completion has gained attention for its ability to generalize across environments without sensor-specific datasets or retraining. However, most existing approaches rely on diffusion-based test-time optimization, which is computationally expensive due to iterative denoising. Recent visual-prompt-based methods reduce training cost but still require repeated forward--backward passes through the full frozen network to optimize input-level prompts, resulting in slow inference. In this work, we show that adapting only the decoder is sufficient for effective test-time optimization, as depth foundation models concentrate depth-relevant information within a low-dimensional decoder subspace. Based on this insight, we propose a lightweight test-time adaptation method that updates only this low-dimensional subspace using sparse depth supervision. Our approach achieves state-of-the-art performance, establishing a new Pareto frontier between accuracy and efficiency for test-time adaptation. Extensive experiments on five indoor and outdoor datasets demonstrate consistent improvements over prior methods, highlighting the practicality of fast zero-shot depth completion.

Doyi Kim, Minseok Seo, Changick Kim

Precipitation forecasting relies on heterogeneous data. Weather radar is accurate, but coverage is geographically limited and costly to maintain. Weather stations provide accurate but sparse point measurements, while satellites offer dense, high-resolution coverage without direct rainfall retrieval. To overcome these limitations, we propose Query-Conditioned Gaussian Splatting (QCGS), the first framework to fuse automatic weather station (AWS) observations with satellite imagery for generating precipitation fields. Unlike conventional 2D Gaussian splatting, which renders the entire image plane, QCGS selectively renders only queried precipitation regions, avoiding unnecessary computation in non-precipitating areas while preserving sharp precipitation structures. The framework combines a radar point proposal network that identifies rainfall-support locations with an implicit neural representation (INR) network that predicts Gaussian parameters for each point. QCGS enables efficient, resolution-flexible precipitation field generation in real time. Through extensive evaluation with benchmark precipitation products, QCGS demonstrates over 50\% improvement in RMSE compared to conventional gridded precipitation products, and consistently maintains high performance across multiple spatiotemporal scales.

John Seon Keun Yi, Minseok Seo, Jongchan Park et al.

Labeling a large set of data is expensive. Active learning aims to tackle this problem by asking to annotate only the most informative data from the unlabeled set. We propose a novel active learning approach that utilizes self-supervised pretext tasks and a unique data sampler to select data that are both difficult and representative. We discover that the loss of a simple self-supervised pretext task, such as rotation prediction, is closely correlated to the downstream task loss. Before the active learning iterations, the pretext task learner is trained on the unlabeled set, and the unlabeled data are sorted and split into batches by their pretext task losses. In each active learning iteration, the main task model is used to sample the most uncertain data in a batch to be annotated. We evaluate our method on various image classification and segmentation benchmarks and achieve compelling performances on CIFAR10, Caltech-101, ImageNet, and Cityscapes. We further show that our method performs well on imbalanced datasets, and can be an effective solution to the cold-start problem where active learning performance is affected by the randomly sampled initial labeled set.

Minseok Seo, Jaemin Lee, Jongchan Park et al.

We propose Sequential Feature Filtering Classifier (FFC), a simple but effective classifier for convolutional neural networks (CNNs). With sequential LayerNorm and ReLU, FFC zeroes out low-activation units and preserves high-activation units. The sequential feature filtering process generates multiple features, which are fed into a shared classifier for multiple outputs. FFC can be applied to any CNNs with a classifier, and significantly improves performances with negligible overhead. We extensively validate the efficacy of FFC on various tasks: ImageNet-1K classification, MS COCO detection, Cityscapes segmentation, and HMDB51 action recognition. Moreover, we empirically show that FFC can further improve performances upon other techniques, including attention modules and augmentation techniques. The code and models will be publicly available.