Hyojin Park

Shubhankar Borse, Hyojin Park, Hong Cai et al.

This paper presents a novel framework to integrate both semantic and instance contexts for panoptic segmentation. In existing works, it is common to use a shared backbone to extract features for both things (countable classes such as vehicles) and stuff (uncountable classes such as roads). This, however, fails to capture the rich relations among them, which can be utilized to enhance visual understanding and segmentation performance. To address this shortcoming, we propose a novel Panoptic, Instance, and Semantic Relations (PISR) module to exploit such contexts. First, we generate panoptic encodings to summarize key features of the semantic classes and predicted instances. A Panoptic Relational Attention (PRA) module is then applied to the encodings and the global feature map from the backbone. It produces a feature map that captures 1) the relations across semantic classes and instances and 2) the relations between these panoptic categories and spatial features. PISR also automatically learns to focus on the more important instances, making it robust to the number of instances used in the relational attention module. Moreover, PISR is a general module that can be applied to any existing panoptic segmentation architecture. Through extensive evaluations on panoptic segmentation benchmarks like Cityscapes, COCO, and ADE20K, we show that PISR attains considerable improvements over existing approaches.

Shubhankar Borse, Debasmit Das, Hyojin Park et al.

We present DejaVu, a novel framework which leverages conditional image regeneration as additional supervision during training to improve deep networks for dense prediction tasks such as segmentation, depth estimation, and surface normal prediction. First, we apply redaction to the input image, which removes certain structural information by sparse sampling or selective frequency removal. Next, we use a conditional regenerator, which takes the redacted image and the dense predictions as inputs, and reconstructs the original image by filling in the missing structural information. In the redacted image, structural attributes like boundaries are broken while semantic context is largely preserved. In order to make the regeneration feasible, the conditional generator will then require the structure information from the other input source, i.e., the dense predictions. As such, by including this conditional regeneration objective during training, DejaVu encourages the base network to learn to embed accurate scene structure in its dense prediction. This leads to more accurate predictions with clearer boundaries and better spatial consistency. When it is feasible to leverage additional computation, DejaVu can be extended to incorporate an attention-based regeneration module within the dense prediction network, which further improves accuracy. Through extensive experiments on multiple dense prediction benchmarks such as Cityscapes, COCO, ADE20K, NYUD-v2, and KITTI, we demonstrate the efficacy of employing DejaVu during training, as it outperforms SOTA methods at no added computation cost.

Debasmit Das, Shubhankar Borse, Hyojin Park et al.

Test-time adaptive (TTA) semantic segmentation adapts a source pre-trained image semantic segmentation model to unlabeled batches of target domain test images, different from real-world, where samples arrive one-by-one in an online fashion. To tackle online settings, we propose TransAdapt, a framework that uses transformer and input transformations to improve segmentation performance. Specifically, we pre-train a transformer-based module on a segmentation network that transforms unsupervised segmentation output to a more reliable supervised output, without requiring test-time online training. To also facilitate test-time adaptation, we propose an unsupervised loss based on the transformed input that enforces the model to be invariant and equivariant to photometric and geometric perturbations, respectively. Overall, our framework produces higher quality segmentation masks with up to 17.6% and 2.8% mIOU improvement over no-adaptation and competitive baselines, respectively.

Kambiz Azarian, Debasmit Das, Hyojin Park et al.

We consider the problem of improving the human instance segmentation mask quality for a given test image using keypoints estimation. We compare two alternative approaches. The first approach is a test-time adaptation (TTA) method, where we allow test-time modification of the segmentation network's weights using a single unlabeled test image. In this approach, we do not assume test-time access to the labeled source dataset. More specifically, our TTA method consists of using the keypoints estimates as pseudo labels and backpropagating them to adjust the backbone weights. The second approach is a training-time generalization (TTG) method, where we permit offline access to the labeled source dataset but not the test-time modification of weights. Furthermore, we do not assume the availability of any images from or knowledge about the target domain. Our TTG method consists of augmenting the backbone features with those generated by the keypoints head and feeding the aggregate vector to the mask head. Through a comprehensive set of ablations, we evaluate both approaches and identify several factors limiting the TTA gains. In particular, we show that in the absence of a significant domain shift, TTA may hurt and TTG show only a small gain in performance, whereas for a large domain shift, TTA gains are smaller and dependent on the heuristics used, while TTG gains are larger and robust to architectural choices.

Hyojin Park, Yi Li, Janghoon Cho et al.

Despite decades of work, surveillance still struggles to find specific targets across long, multi-camera video. Prior methods -- tracking pipelines, CLIP based models, and VideoRAG -- require heavy manual filtering, capture only shallow attributes, and fail at temporal reasoning. Real-world searches are inherently multimodal (e.g., "When does this person join the fight?" with the person's image), yet this setting remains underexplored. Also, there are no proper benchmarks to evaluate those setting - asking video with multimodal queries. To address this gap, we introduce ForeSeaQA, a new benchmark specifically designed for video QA with image-and-text queries and timestamped annotations of key events. The dataset consists of long-horizon surveillance footage paired with diverse multimodal questions, enabling systematic evaluation of retrieval, temporal grounding, and multimodal reasoning in realistic forensic conditions. Not limited to this benchmark, we propose ForeSea, an AI forensic search system with a 3-stage, plug-and-play pipeline. (1) A tracking module filters irrelevant footage; (2) a multimodal embedding module indexes the remaining clips; and (3) during inference, the system retrieves top-K candidate clips for a Video Large Language Model (VideoLLM) to answer queries and localize events. On ForeSeaQA, ForeSea improves accuracy by 3.5% and temporal IoU by 11.0 over prior VideoRAG models. To our knowledge, ForeSeaQA is the first benchmark to support complex multimodal queries with precise temporal grounding, and ForeSea is the first VideoRAG system built to excel in this setting.

Hyojin Park, Jayeon Yoo, Seohyeong Jeong et al.

Current state-of-the-art approaches for Semi-supervised Video Object Segmentation (Semi-VOS) propagates information from previous frames to generate segmentation mask for the current frame. This results in high-quality segmentation across challenging scenarios such as changes in appearance and occlusion. But it also leads to unnecessary computations for stationary or slow-moving objects where the change across frames is minimal. In this work, we exploit this observation by using temporal information to quickly identify frames with minimal change and skip the heavyweight mask generation step. To realize this efficiency, we propose a novel dynamic network that estimates change across frames and decides which path -- computing a full network or reusing previous frame's feature -- to choose depending on the expected similarity. Experimental results show that our approach significantly improves inference speed without much accuracy degradation on challenging Semi-VOS datasets -- DAVIS 16, DAVIS 17, and YouTube-VOS. Furthermore, our approach can be applied to multiple Semi-VOS methods demonstrating its generality. The code is available in https://github.com/HYOJINPARK/Reuse_VOS.

Hyojin Park, Ganesh Venkatesh, Nojun Kwak

Semi-supervised video object segmentation (semi-VOS) is widely used in many applications. This task is tracking class-agnostic objects from a given target mask. For doing this, various approaches have been developed based on online-learning, memory networks, and optical flow. These methods show high accuracy but are hard to be utilized in real-world applications due to slow inference time and tremendous complexity. To resolve this problem, template matching methods are devised for fast processing speed but sacrificing lots of performance in previous models. We introduce a novel semi-VOS model based on a template matching method and a temporal consistency loss to reduce the performance gap from heavy models while expediting inference time a lot. Our template matching method consists of short-term and long-term matching. The short-term matching enhances target object localization, while long-term matching improves fine details and handles object shape-changing through the newly proposed adaptive template attention module. However, the long-term matching causes error-propagation due to the inflow of the past estimated results when updating the template. To mitigate this problem, we also propose a temporal consistency loss for better temporal coherence between neighboring frames by adopting the concept of a transition matrix. Our model obtains 79.5% J&F score at the speed of 73.8 FPS on the DAVIS16 benchmark. The code is available in https://github.com/HYOJINPARK/TTVOS.

Hyojin Park, Lars Lowe Sjösund, YoungJoon Yoo et al.

Designing a lightweight and robust portrait segmentation algorithm is an important task for a wide range of face applications. However, the problem has been considered as a subset of the object segmentation problem and less handled in the semantic segmentation field. Obviously, portrait segmentation has its unique requirements. First, because the portrait segmentation is performed in the middle of a whole process of many real-world applications, it requires extremely lightweight models. Second, there has not been any public datasets in this domain that contain a sufficient number of images with unbiased statistics. To solve the first problem, we introduce the new extremely lightweight portrait segmentation model SINet, containing an information blocking decoder and spatial squeeze modules. The information blocking decoder uses confidence estimates to recover local spatial information without spoiling global consistency. The spatial squeeze module uses multiple receptive fields to cope with various sizes of consistency in the image. To tackle the second problem, we propose a simple method to create additional portrait segmentation data which can improve accuracy on the EG1800 dataset. In our qualitative and quantitative analysis on the EG1800 dataset, we show that our method outperforms various existing lightweight segmentation models. Our method reduces the number of parameters from 2.1M to 86.9K (around 95.9% reduction), while maintaining the accuracy under an 1% margin from the state-of-the-art portrait segmentation method. We also show our model is successfully executed on a real mobile device with 100.6 FPS. In addition, we demonstrate that our method can be used for general semantic segmentation on the Cityscapes dataset. The code and dataset are available in https://github.com/HYOJINPARK/ExtPortraitSeg .

Hyojin Park, Lars Lowe Sjösund, YoungJoon Yoo et al.

Designing a lightweight and robust portrait segmentation algorithm is an important task for a wide range of face applications. However, the problem has been considered as a subset of the object segmentation problem. bviously, portrait segmentation has its unique requirements. First, because the portrait segmentation is performed in the middle of a whole process of many realworld applications, it requires extremely lightweight models. Second, there has not been any public datasets in this domain that contain a sufficient number of images with unbiased statistics. To solve the problems, we introduce a new extremely lightweight portrait segmentation model consisting of a two-branched architecture based on the concentrated-comprehensive convolutions block. Our method reduces the number of parameters from 2.1M to 37.7K (around 98.2% reduction), while maintaining the accuracy within a 1% margin from the state-of-the-art portrait segmentation method. In our qualitative and quantitative analysis on the EG1800 dataset, we show that our method outperforms various existing lightweight segmentation models. Second, we propose a simple method to create additional portrait segmentation data which can improve accuracy on the EG1800 dataset. Also, we analyze the bias in public datasets by additionally annotating race, gender, and age on our own. The augmented dataset, the additional annotations and code are available in https://github.com/HYOJINPARK/ExtPortraitSeg .

Byeongho Heo, Jeesoo Kim, Sangdoo Yun et al.

We investigate the design aspects of feature distillation methods achieving network compression and propose a novel feature distillation method in which the distillation loss is designed to make a synergy among various aspects: teacher transform, student transform, distillation feature position and distance function. Our proposed distillation loss includes a feature transform with a newly designed margin ReLU, a new distillation feature position, and a partial L2 distance function to skip redundant information giving adverse effects to the compression of student. In ImageNet, our proposed method achieves 21.65% of top-1 error with ResNet50, which outperforms the performance of the teacher network, ResNet152. Our proposed method is evaluated on various tasks such as image classification, object detection and semantic segmentation and achieves a significant performance improvement in all tasks. The code is available at https://sites.google.com/view/byeongho-heo/overhaul

Shubhankar Borse, Kartikeya Bhardwaj, Mohammad Reza Karimi Dastjerdi et al.

Diffusion models are increasingly popular for generative tasks, including personalized composition of subjects and styles. While diffusion models can generate user-specified subjects performing text-guided actions in custom styles, they require fine-tuning and are not feasible for personalization on mobile devices. Hence, tuning-free personalization methods such as IP-Adapters have progressively gained traction. However, for the composition of subjects and styles, these works are less flexible due to their reliance on ControlNet, or show content and style leakage artifacts. To tackle these, we present SubZero, a novel framework to generate any subject in any style, performing any action without the need for fine-tuning. We propose a novel set of constraints to enhance subject and style similarity, while reducing leakage. Additionally, we propose an orthogonalized temporal aggregation scheme in the cross-attention blocks of denoising model, effectively conditioning on a text prompt along with single subject and style images. We also propose a novel method to train customized content and style projectors to reduce content and style leakage. Through extensive experiments, we show that our proposed approach, while suitable for running on-edge, shows significant improvements over state-of-the-art works performing subject, style and action composition.

Shweta Mahajan, Hoang Le, Hyojin Park et al.

Large Vision-Language Models (VLMs) rely on effective multimodal alignment between pre-trained vision encoders and Large Language Models (LLMs) to integrate visual and textual information. This paper presents a comprehensive analysis of attention patterns in efficient VLMs, revealing that concatenation-based architectures frequently fail to distinguish between semantically matching and non-matching image-text pairs. This is a key factor for object hallucination in these models. To address this, we introduce Attention-Guided Efficient Vision-Language Models (AGE-VLM), a novel framework that enhances visual grounding through interleaved cross-attention layers to instill vision capabilities in pretrained small language models. This enforces in VLM the ability "look" at the correct image regions by leveraging spatial knowledge distilled from the Segment Anything Model (SAM), significantly reducing hallucination. We validate our approach across different vision-centric benchmarks where our method is better or comparable to prior work on efficient VLMs. Our findings provide valuable insights for future research aimed at achieving enhanced visual and linguistic understanding in VLMs.

Jungsoo Lee, Janghoon Cho, Hyojin Park et al.

Despite their consistent performance improvements, cross-modal retrieval models (e.g., CLIP) show degraded performances with retrieving keys composed of fused image-text modality (e.g., Wikipedia pages with both images and text). To address this critical challenge, multimodal retrieval has been recently explored to develop a unified single retrieval model capable of retrieving keys across diverse modality combinations. A common approach involves constructing new composed sets of image-text triplets (e.g., retrieving a pair of image and text given a query image). However, such an approach requires careful curation to ensure the dataset quality and fails to generalize to unseen modality combinations. To overcome these limitations, this paper proposes Generalized Contrastive Learning (GCL), a novel loss formulation that improves multimodal retrieval performance without the burdensome need for new dataset curation. Specifically, GCL operates by enforcing contrastive learning across all modalities within a mini-batch, utilizing existing image-caption paired datasets to learn a unified representation space. We demonstrate the effectiveness of GCL by showing consistent performance improvements on off-the-shelf multimodal retrieval models (e.g., VISTA, CLIP, and TinyCLIP) using the M-BEIR, MMEB, and CoVR benchmarks.

Minho Park, Sunghyun Park, Jungsoo Lee et al.

This paper addresses the challenge of data scarcity in semantic segmentation by generating datasets through text-to-image (T2I) generation models, reducing image acquisition and labeling costs. Segmentation dataset generation faces two key challenges: 1) aligning generated samples with the target domain and 2) producing informative samples beyond the training data. Fine-tuning T2I models can help generate samples aligned with the target domain. However, it often overfits and memorizes training data, limiting their ability to generate diverse and well-aligned samples. To overcome these issues, we propose Concept-Aware LoRA (CA-LoRA), a novel fine-tuning approach that selectively identifies and updates only the weights associated with necessary concepts (e.g., style or viewpoint) for domain alignment while preserving the pretrained knowledge of the T2I model to produce informative samples. We demonstrate its effectiveness in generating datasets for urban-scene segmentation, outperforming baseline and state-of-the-art methods in in-domain (few-shot and fully-supervised) settings, as well as in domain generalization tasks, especially under challenging conditions such as adverse weather and varying illumination, further highlighting its superiority.

Vibashan VS, Shubhankar Borse, Hyojin Park et al.

In this paper, we introduce an open-vocabulary panoptic segmentation model that effectively unifies the strengths of the Segment Anything Model (SAM) with the vision-language CLIP model in an end-to-end framework. While SAM excels in generating spatially-aware masks, it's decoder falls short in recognizing object class information and tends to oversegment without additional guidance. Existing approaches address this limitation by using multi-stage techniques and employing separate models to generate class-aware prompts, such as bounding boxes or segmentation masks. Our proposed method, PosSAM is an end-to-end model which leverages SAM's spatially rich features to produce instance-aware masks and harnesses CLIP's semantically discriminative features for effective instance classification. Specifically, we address the limitations of SAM and propose a novel Local Discriminative Pooling (LDP) module leveraging class-agnostic SAM and class-aware CLIP features for unbiased open-vocabulary classification. Furthermore, we introduce a Mask-Aware Selective Ensembling (MASE) algorithm that adaptively enhances the quality of generated masks and boosts the performance of open-vocabulary classification during inference for each image. We conducted extensive experiments to demonstrate our methods strong generalization properties across multiple datasets, achieving state-of-the-art performance with substantial improvements over SOTA open-vocabulary panoptic segmentation methods. In both COCO to ADE20K and ADE20K to COCO settings, PosSAM outperforms the previous state-of-the-art methods by a large margin, 2.4 PQ and 4.6 PQ, respectively. Project Website: https://vibashan.github.io/possam-web/.

Hyojin Park, Youngjoon Yoo, Geonseok Seo et al.

One of the practical choices for making a lightweight semantic segmentation model is to combine a depth-wise separable convolution with a dilated convolution. However, the simple combination of these two methods results in an over-simplified operation which causes severe performance degradation due to loss of information contained in the feature map. To resolve this problem, we propose a new block called Concentrated-Comprehensive Convolution (C3) which applies the asymmetric convolutions before the depth-wise separable dilated convolution to compensate for the information loss due to dilated convolution. The C3 block consists of a concentration stage and a comprehensive convolution stage. The first stage uses two depth-wise asymmetric convolutions for compressed information from the neighboring pixels to alleviate the information loss. The second stage increases the receptive field by using a depth-wise separable dilated convolution from the feature map of the first stage. We applied the C3 block to various segmentation frameworks (ESPNet, DRN, ERFNet, ENet) for proving the beneficial properties of our proposed method. Experimental results show that the proposed method preserves the original accuracies on Cityscapes dataset while reducing the complexity. Furthermore, we modified ESPNet to achieve about 2% better performance while reducing the number of parameters by half and the number of FLOPs by 35% compared with the original ESPNet. Finally, experiments on ImageNet classification task show that C3 block can successfully replace dilated convolutions.

Hyojin Park, YoungJoon Yoo, Nojun Kwak

In this paper, we introduce a new method for generating an object image from text attributes on a desired location, when the base image is given. One step further to the existing studies on text-to-image generation mainly focusing on the object's appearance, the proposed method aims to generate an object image preserving the given background information, which is the first attempt in this field. To tackle the problem, we propose a multi-conditional GAN (MC-GAN) which controls both the object and background information jointly. As a core component of MC-GAN, we propose a synthesis block which disentangles the object and background information in the training stage. This block enables MC-GAN to generate a realistic object image with the desired background by controlling the amount of the background information from the given base image using the foreground information from the text attributes. From the experiments with Caltech-200 bird and Oxford-102 flower datasets, we show that our model is able to generate photo-realistic images with a resolution of 128 x 128. The source code of MC-GAN is released.

Hyojin Park, Jisoo Jeong, Youngjoon Yoo et al.

Semantic segmentation, like other fields of computer vision, has seen a remarkable performance advance by the use of deep convolution neural networks. However, considering that neighboring pixels are heavily dependent on each other, both learning and testing of these methods have a lot of redundant operations. To resolve this problem, the proposed network is trained and tested with only 0.37% of total pixels by superpixel-based sampling and largely reduced the complexity of upsampling calculation. The hypercolumn feature maps are constructed by pyramid module in combination with the convolution layers of the base network. Since the proposed method uses a very small number of sampled pixels, the end-to-end learning of the entire network is difficult with a common learning rate for all the layers. In order to resolve this problem, the learning rate after sampling is controlled by statistical process control (SPC) of gradients in each layer. The proposed method performs better than or equal to the conventional methods that use much more samples on Pascal Context, SUN-RGBD dataset.

Jisoo Jeong, Hyojin Park, Nojun Kwak

We propose an object detection method that improves the accuracy of the conventional SSD (Single Shot Multibox Detector), which is one of the top object detection algorithms in both aspects of accuracy and speed. The performance of a deep network is known to be improved as the number of feature maps increases. However, it is difficult to improve the performance by simply raising the number of feature maps. In this paper, we propose and analyze how to use feature maps effectively to improve the performance of the conventional SSD. The enhanced performance was obtained by changing the structure close to the classifier network, rather than growing layers close to the input data, e.g., by replacing VGGNet with ResNet. The proposed network is suitable for sharing the weights in the classifier networks, by which property, the training can be faster with better generalization power. For the Pascal VOC 2007 test set trained with VOC 2007 and VOC 2012 training sets, the proposed network with the input size of 300 x 300 achieved 78.5% mAP (mean average precision) at the speed of 35.0 FPS (frame per second), while the network with a 512 x 512 sized input achieved 80.8% mAP at 16.6 FPS using Nvidia Titan X GPU. The proposed network shows state-of-the-art mAP, which is better than those of the conventional SSD, YOLO, Faster-RCNN and RFCN. Also, it is faster than Faster-RCNN and RFCN.