Suhwan Cho

I am a Ph.D. student at Yonsei University, Seoul, Korea. I was a Research Scientist Intern at Adobe Research, San Jose, California, USA.

My research mainly focuses on video editing technologies such as video object segmentation and video inpainting.

I am always open to collaborations! Please feel free to contact me if you are interested :)

CV / GitHub / Linkedin / Google Scholar

We propose two novel prototype-based attention mechanisms to incorporate different modalities and frames via dense propagation across them.

We propose a guided slot attention network to reinforce spatial structural information and obtain better foreground–background separation.

We propose the inter-frame curve network to effectively leverage the spatio-temporal dependency of the human skeleton.

We propose a novel network architecture that operates regardless of motion availability, termed as a motion-as-option network. A collaborative network learning strategy is also introduced to fully exploit the property of the proposed network.

We propose a prototype memory network architecture to effectively extract and store appearance and motion information based on component prototypes from RGB images and optical flow maps.

We propose three novel strategies to suppress the negative influence of background distractions in semi-supervised video object segmentation.

We propose a prototype sampling network that only samples prototypes corresponding to obtain robustness to inconsistencies between RGB images and depth maps.

We introduce a bijective matching mechanism to find the best matches from the query frame to the reference frame and also vice versa. A mask embedding module is also proposed to consider multiple historic masks simultaneously.

We propose two novel prototype-based attention mechanisms to incorporate different modalities and frames via dense propagation across them.

We propose a guided slot attention network to reinforce spatial structural information and obtain better foreground–background separation.

We propose the inter-frame curve network to effectively leverage the spatio-temporal dependency of the human skeleton.

We propose an adaptive graph convolution module to capture the local structural information and long-range dependencies between distant pixels effectively.

We propose a novel framework for unsupervised video object segmentation, which can utilize both contextual and motion information from adjacent frames.

We propose a two-stream decoder network, designed to learn both normal and abnormal features. A feature normality estimator is also proposed to eliminate abnormal features and prevent high-quality reconstruction of abnormal regions.

We propose fast adaptive patch memory consisting of patch-wise and layer-wise memory banks. A patch-wise adaptive coreset sampling is also proposed for fast and accurate detection.

We propose a novel network architecture that operates regardless of motion availability, termed as a motion-as-option network. A collaborative network learning strategy is also introduced to fully exploit the property of the proposed network.

We propose a prototype memory network architecture to effectively extract and store appearance and motion information based on component prototypes from RGB images and optical flow maps.

We propose three novel strategies to suppress the negative influence of background distractions in semi-supervised video object segmentation.

We propose a prototype sampling network that only samples prototypes corresponding to obtain robustness to inconsistencies between RGB images and depth maps.

We propose a detection-identification balancing margin loss for minimizing the adverse effects caused by two different objectives in one-stage multi-object tracking.

We propose a superpixel group-correlation network architecture that uses a superpixel algorithm to obtain various component features from a group of images and creates a group-correlation matrix to detect the common components of those images.

We propose a novel occlusion correction network that corrects features through relational weight learning and obtains diverse and representative features without using external networks.

We propose an implicit two-path AE, a structure in which two encoders implicitly model appearance and motion features, along with a single decoder that combines them to learn normal video patterns.

We introduce a bijective matching mechanism to find the best matches from the query frame to the reference frame and also vice versa. A mask embedding module is also proposed to consider multiple historic masks simultaneously.

We propose a simple and effective lightweight graph convolutional network for skeleton-based action recognition.

We propose a real-time clue refining network for video object segmentation that does not have any intermediate network to efficiently deal with simple scenarios.

We extend video inpainting to one-shot video inpainting, which refers to erasing a designated object in a video sequence only using a single frame annotation.

We propose deformable point sampling method and global-local aggregation architecture to integrate object's global information, background, and boundary local information to improve the camouflaged object detection.

We propose an equalized matching mechanism to overcome the limitations of existing matching methods for video object segmentation.

This website's source code is borrowed from jonbarron's website.