HoloQRam: Efficient Real-Time Spatial Video Delivery via Animated QR Codes

This is the official implementation of HoloQRam: Efficient Real-Time Spatial Video Delivery via Animated QR Codes.

Jaewon Choi and JeongGil Ko

Abstract

HoloQRam enables real-time, infrastructure-free spatial video delivery using animated QR codes on commodity displays with mobile RGB-D reconstruction. The system transmits lightweight 128x128 RGB video streams encoded as sequential QR symbols on standard displays, uses a Genetic Algorithm (GA)-based image filtering pipeline to improve decoding robustness under motion blur, reflections, and lighting variability, and employs a hybrid Transformer-CNN model that jointly performs 4x super-resolution and depth estimation on the receiver side. HoloQRam achieves up to 70% higher decoding success rates than baselines under adverse conditions, real-time throughput >30 fps, and 30.8 dB PSNR reconstruction quality, enabling spontaneous one-to-many spatial video broadcasting without dedicated infrastructure.

About This Repository

This repository contains the receiver-side reconstruction pipeline (Stage 2), including:

• Dataset preparation: C++ point cloud renderer generating multi-view RGB-D training pairs from the 8iVFB dataset
• Model training: Hybrid Transformer-CNN for joint 4x super-resolution (128x128 → 512x512) and monocular depth estimation
• Evaluation: Quantitative assessment of SR and depth reconstruction quality

System Requirements

C++ Build Dependencies (apt-get)

sudo apt-get update
sudo apt-get install -y \
    build-essential \
    cmake \
    pkg-config \
    libgl1-mesa-dev \
    libegl1-mesa-dev \
    libgles2-mesa-dev \
    libturbojpeg0-dev \
    libglm-dev

Package	Description
build-essential	GCC, G++, make
cmake	Build system (>= 3.16)
pkg-config	Package configuration tool
libgl1-mesa-dev	OpenGL development libraries
libegl1-mesa-dev	EGL development libraries
libgles2-mesa-dev	OpenGL ES development libraries
libturbojpeg0-dev	libjpeg-turbo for fast JPEG encoding
libglm-dev	OpenGL Mathematics library (header-only)

Python Dependencies

pip install -r scripts/requirements.txt

Requires Python 3.8+ and CUDA-compatible PyTorch.

Building C++ Tools

cd scripts/prepare_dataset_cpp
mkdir -p build && cd build
cmake ..
make -j$(nproc)

The executable will be at build/prepare_dataset.

Project Structure

scripts/
├── prepare_dataset_cpp/    # C++ point cloud renderer
│   ├── src/                # Source files
│   ├── include/            # Header files
│   └── CMakeLists.txt
├── train.py                # Training script
├── evaluate.py             # Evaluation script
├── prepare_dataset.py      # Python dataset preparation
└── requirements.txt        # Python dependencies