Projects

Current Projects

Threshold Matters in WSSS: Manipulating the Activation for the Robust and Accurate Segmentation Model Against Thresholds

Weakly supervised semantic segmentation technique for robustness against the choice of the global threshold.

Pretraining GANs with Synthetic Dataset



Improving low-shot image generation with GANs via pretraining with synthetic datasets consisting of primitive shapes such as circles, rectangle,s and lines.

Railroad is not a Train: Saliency as Pseudo-pixel Supervision for Weakly Supervised Semantic Segmentation

Weakly supervised semantic segmentation technique for convolutional neural networks

Few-shot Font Generation


Reconstructing full font library only with few reference samples, e.g., 8.

GridMix: Strong regularization through local context mapping


Regularization technique for convolutional neural networks

Evaluating Weakly Supervised Object Localization Right


Fair and realistic evaluation framework for weakly supervised object localization

Attention-based Dropout Layer for Weakly Supervised Object Localization


Localizing full extent of the object without pixel-level annotations

Resembled GANs


Simultaneously generating two domain images with similar attributes
https://sites.google.com/site/cvgyonsei/projects/Thumbnail_ICML.png

Unsupervised Object Localization


Unsupervised object localization through self-supervised approach

Editing and Generating Facial Images using GAN


Editing the attributes of existing images or generating images with desired attributes simultaneously through single training procedure
3D Reconstruction

Depth Reconstruction of Translucent Objects from single ToF camera


Fully automated, 1-2 shot depth correction with only Kinect V2 depth camera and a Resnet

Rigid and Non-rigid Motion Artifact Reduction in X-ray CT


CT motion artifact reduction using attention module in real-time

Completed Projects

Inverse Lighting


Extracting the lighting information from a pair of color and depth image
https://sites.google.com/site/cvgyonsei/projects/facerecognition

Face Recognition


Low-shot learning: How to train your network when there is not enough data?

Recommender System With Visual Feature


Employing visual features of movie posters to enhance the performance of recommender system

Environment Map From a Single Image


Predicting the surrounding environment map from a single image


3D Motion Artifact Compensation in CT with Depth Camera


A new motion-free CT system to handle personalized motions using depth cameras
https://sites.google.com/site/cvgyonsei/projects/transform_photograph

Aesthetic Image Enhancement

Depth Reconstruction of Translucent Objects


Reconstructing a three dimensional (3-D) translucent object using a single time-of-flight depth camera with simple user marks

Depth Image Enhancement using Perceptual Texture Priors


A perceptual-based technique by restoring the geometric details

Face Detection


Finding and classifying Face

Semantic Aware Neural Style Transfer


A style transfer network with semantic awareness