Beomyoung Kim 김범영

A diagnostic framework for 3D spatial reasoning that turns silent perception failures in visual program synthesis into typed diagnoses, driving targeted program repair to rival frontier VLMs without task-specific training.

A segmentation mask-refinement framework that synthesizes realistic, semantic-aware errors via adversarial perturbation and corrects them with contrastive learning, consistently improving state-of-the-art segmentation models.

ZIM: Zero-Shot Image Matting for Anything

@inproceedings{kim2025zim,
  title={ZIM: Zero-Shot Image Matting for Anything},
  author={Kim, Beomyoung and others},
  booktitle={ICCV}, year={2025}}

Towards Label-Efficient Human Matting

Rethinking Saliency-Guided Weakly-Supervised Semantic Segmentation

ECLIPSE: Efficient Continual Learning in Panoptic Segmentation with Visual Prompt Tuning

EResFD: Rediscovery of the Effectiveness of Standard Convolution for Lightweight Face Detection

The Devil is in the Points: Weakly Semi-Supervised Instance Segmentation via Point-Guided Mask Representation

Beyond Semantic to Instance Segmentation via Semantic Knowledge Transfer and Self-Refinement

Learning Features with Parameter-Free Layers

TricubeNet: 2D Kernel-Based Object Representation for Weakly-Occluded Oriented Object Detection

SSUL: Semantic Segmentation with Unknown Label for Exemplar-based Class-Incremental Learning

Discriminative Region Suppression for Weakly-Supervised Semantic Segmentation

3D Point Cloud Upsampling and Colorization using GAN

Fully Automated Valet Parking System Based on Infrastructure Sensing