Skills

Research Expertise

Generative Models

My diffusion-model research focuses on rethinking design choices such as spatial granularity, conditioning structure, and frequency representation. I have implemented DDPM, DDIM, and EDM from scratch in both Python and C++, and I am experienced in redesigning network architectures, training generative models at scale, and modifying the reverse process.

Computer Vision

I have spent six years doing research at a CV/CG laboratory, covering image generation, synthetic pre-training, metric learning, and image retrieval. I am familiar with standard CV pipelines, from dataset construction and large-scale training to evaluation with generative-quality and retrieval metrics (FID, LPIPS, Recall@K).

Temporal / Video Generation

During my visit to Aalto University, I contributed to a collaborator’s research on time-lapse image generation with conditional diffusion models. I focused on architectural modifications that address the abrupt visual jumps arising when the conditioning signal is varied smoothly, a structural challenge directly tied to temporal consistency in generated content.

3D Vision & Graphics

I have hands-on working knowledge of NeRF and COLMAP. At Insight Inc., I have handled 3D meshes in practice for several years, including mesh preparation for finite-element solvers and visualization of simulation results in ParaView. I also have introductory familiarity with Vulkan, Blender, Maya, and Houdini.

Interactive Research Tooling

I believe that interactive exploration of experimental results is essential for forming research insights, and I have built custom visualization tools throughout my past projects to support this practice. When off-the-shelf visualization tools are insufficient, I am comfortable building real-time interactive interfaces with OpenGL and ImGui from scratch.

fft_viz: an ImGui-based FFT visualizer I made for analyzing the frequency-domain behavior of images during diffusion-model research.

Technical Skills

• Programming Languages: Python, C, C++, CUDA, and Java (all proficient; experienced in developing codebases on the order of tens of thousands of lines)
• Deep Learning Frameworks: PyTorch (primary), TensorFlow (proficient)
• Distributed Training & Infrastructure: Distributed Data Parallel (DDP, frequently used), Slurm, Docker, Singularity, supercomputers including Fugaku and ABCI, and Weights & Biases (wandb)