Video Frame Interpolation (VFI) Engine

This repository provides three high-performance engines for increasing video frame rates (FPS) and achieving "butter-smooth" playback.

• Best Engine (Python): Our premium engine featuring TTA and CRF 0 for mathematically lossless results.
• Standard Engine (Python): Uses Deep Learning (RIFE/IFNet) for high visual quality.
• C++ Engine: Uses Multi-Scale Pyramid Motion Estimation for high-speed processing without requiring a GPU.

---

Quick Comparison

Feature	Best Engine	Standard Engine	C++ Engine
Technology	DL + TTA + Lossless	DL (RIFE)	Pyramid Flow
Quality	⭐⭐⭐⭐⭐+ (Absolute Best)	⭐⭐⭐⭐⭐ (Excellent)	⭐⭐⭐ (Good / Fast)
Speed	Slowest	Faster	Fastest
Specialty	0 Quality Loss	High Fidelity	Real-time / Low-power

---

1. Best Engine (Python) - py/best_engine.py

This is the ultimate engine for users who prioritize quality above all else.

Key Features:

• Test Time Augmentation (TTA): Runs multiple passes per frame to eliminate artifacts.
• Mathematically Lossless: Supports --lossless (CRF 0) encoding.
• Scene Detection: Automatically prevents warping at camera cuts.
• Async I/O: Multi-threaded video reading and writing.

Usage

# Maximum Quality Mode
python py/best_engine.py --video in.mp4 --output out.mp4 --fps 60 --tta --lossless

# Balanced Mode
python py/best_engine.py --video in.mp4 --output out.mp4 --fps 60 --batch_size 4

---

2. Standard Engine (Python) - py/main.py

Uses the RIFE (Real-Time Intermediate Flow Estimation) architecture.

Usage

python py/main.py --video input.mp4 --output output.mp4 --fps 60

---

3. C++ Engine (High Performance)

A custom implementation of a Multi-Scale Pyramid Motion Estimator.

Build

cd cpp && mkdir -p build && cd build
cmake .. && make -j$(nproc)

Usage

./vfi <input_video> <output_video> <target_fps>

---

🛠 Prerequisites

• FFmpeg: Required for video encoding/decoding.
• Python 3.8+
• CUDA: Recommended for Python engines.
• CMake & G++: For building the C++ engine.

---

📁 Project Structure

• py/best_engine.py: Advanced premium engine.
• py/main.py: Standard RIFE implementation.
• cpp/src/: High-performance motion estimation core (C++).