Projects

---

Tarka Embedding V1

The Tarka Embedding V1 series is a compact and efficient text embedding model family developed to explore the capabilities of knowledge distillation, coreset selection, and model compression techniques

Open - Source

OM-NeRF

We present OM-NeRF, a method for comprehensive 3D scene manipulation built on Zip-NeRF. It employs three prediction heads — Composited Scene, Background, and Semantic — to predict RGB values, background content, and object codes, respectively

VAGI AI

VAGI (Vision AI for General Intelligence) aims to move multimodal large language models (MLLMs) beyond simple QA and summarization, enabling deeper visual reasoning, effective tool use, and adaptive strategies based on context and input.

Open - Source

MLT-STARK

MLT-STARK (Memory-based Long-Term STARK)is a memory-augmented STARK tracker for anti-UAV tracking, using historical cues to improve target re-identification, enabling adaptive responses to occlusions and sudden motion shifts.

Enhanced Deep OC-SORT

A memory-augmented Deep OC-SORT tracker that leverages stored historical track information and a novel similarity measure to improve object re-identification, enabling more reliable ID consistency under occlusion, appearance similarity, and object disappearance and reappearance.