Skip to content
View JayBuendia's full-sized avatar

Block or report JayBuendia

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
JayBuendia/README.md

Renjie Shen

3D Vision · Point Clouds · Language-Guided Learning · LLM Systems

Building robust learning systems for 3D perception and intelligent agents.

while (curious) { learn(); build(); verify(); }


About

I am a researcher working at the intersection of 3D vision and language-guided learning, with a focus on point cloud understanding, few-shot generalization, and agent distillation.

Current focus: using language priors, adaptive supervision, and on-policy data to improve model robustness under limited annotation and distribution shift.

Featured Research

Runnable C++20 experiments that turn LLM systems questions into numerical checks: RoPE, KV cache, online softmax, attention layouts, speculative decoding, and tiled GEMM.

C++20 LLM Inference Systems From Scratch

Recoverability-aware adaptation and gated distillation for few-shot outdoor point cloud segmentation.

Few-shot Learning 3D Segmentation Distillation

Language priors and text prototypes for semantic segmentation with pretrained 3D backbones.

Language Guidance Point Clouds CLIP

On-policy relabeling for robust search-agent distillation under rollout distribution shift.

LLM Agents On-policy Learning LoRA

Local-global feature learning for large-scale indoor point cloud semantic segmentation.

Indoor Scenes Local-Global Learning C++

Research Focus

  • 3D scene understanding: semantic segmentation of indoor and outdoor point clouds
  • Learning with limited supervision: few-shot, open-vocabulary, and distillation-based learning
  • Language-guided perception: connecting text priors with pretrained 3D representations
  • Agent learning: policy distillation and on-policy data collection for search agents
  • LLM systems: readable C++ experiments for inference algorithms and memory behavior

Toolbox

Python PyTorch CUDA C++ Git

Background

My academic background includes Changzhou University and the University of Reading.

Pinned Loading

  1. LangPoint LangPoint Public

    Language-guided point cloud segmentation with text prototypes and pretrained 3D backbones.

    Python 1

  2. SearchAgent-OPD SearchAgent-OPD Public

    On-policy relabeling for robust search-agent distillation under rollout distribution shift.

    Python 1

  3. LG-Net LG-Net Public

    Local-global feature learning for large-scale indoor point cloud semantic segmentation.

    C++

  4. Recover3D Recover3D Public

    Recoverability-aware adaptation and gated distillation for few-shot outdoor point cloud segmentation.

    Python

  5. LLMCore-CPP LLMCore-CPP Public

    Runnable C++20 experiments for RoPE, KV cache, speculative decoding, attention layouts, online softmax, and GEMM.

    C++