ReForge Compression Tool

ReForge Compression Tool

Significant Game Size Reduction with Two Innovative Methods: Model and Texture Reconstruction

Author: Reza Basafa (Age 16)

Introduction :

In modern games, the massive size of data is one of the primary challenges. Most of this size comes from 3D models with millions of vertices and high-resolution textures. Reducing this size without noticeable quality loss is the key to smoother user experience and faster game downloads.

This paper introduces two innovative methods:

Reconstruction of 3D models with Machine Learning (ML) during installation

Texture reconstruction using Base Texture and Recipt during installation

Both approaches are designed to significantly reduce game size while enabling fast loading on GPU and CPU.

Section 1: 3D Model Reconstruction with Machine Learning
1.1 Challenges

Large models: 1,000,000 vertices or more

Typical size: 12 MB

With RGB8 textures: ≈ 3 MB

1.2 Compression Method

a) Channel Combination
Instead of three separate textures for X, Y, Z, one RGB texture is used:

R → X

G → Y

B → Z

This reduces file count and loading time.

b) Quantization
Convert coordinates from float32 to uint16 or uint8:

float32 precision = 1e-7

uint16 precision ≈ 1e-4 → acceptable for most models

Reduces size without significant geometric distortion

c) Delta Encoding

Store differences between consecutive vertices

Normalize differences to 0–1 and store in texture

GPU-friendly compression using ASTC and BC3 formats

Fast decode on GPU, final size reduced up to 90%

d) Model Reconstruction with ML

Vertex coordinates are extracted from 2D textures generated from 3D models

A hybrid Deep Learning model analyzes:

How vertices connect to form edges

How edges connect to form faces

Output: reconstructed 3D model, smaller in size yet close to the original geometry, ready for GPU loading

1.3 Example Result
Model Original Size After Compression
1,000,000 vertices 12 MB 1–1.5 MB

≈ 90% reduction

1.4 Advantages and Limitations

Advantages:

Significant size reduction

Direct GPU loading

Can be combined with LOD and Progressive Loading

Limitations:

Too much quantization may cause slight geometric shifts

Vertex counts must be convertible to 2D textures

Section 2: Texture Reconstruction with Base Texture and Recipt
2.1 Challenges

Textures represent the largest portion of game size

High-resolution textures can reach tens of gigabytes

2.2 Reconstruction Process

Step 1: Pre-processing

All textures are converted into Base Texture + Recipt

Base Textures: compressed, small-scale textures used for reconstruction

Recipt: instruction file for rebuilding final textures, including:

Base Color

Masks, Roughness, Metallic, AO, etc.

Normals

Height Map

Other details required for high-quality reconstruction

Step 2: Game Installation

Final textures are rebuilt based on Recipt

Initial download size significantly reduced

2.3 Example Result

Texture Original Size Base + Recipt Size
Single Texture Pack 3 GB 300 MB

≈ 90% download reduction

2.4 Advantages and Limitations

Advantages:

Significant reduction in download size

High-quality texture reconstruction

Easy customization and updates

Limitations:

Requires specialized software to generate Recipt

Installation may take longer (up to 30 minutes on mid/high-end systems)

Complex layer and mask management

Section 3: Target Hardware and Installation Predictions

Mid/high-end systems: installation overhead up to 30 minutes

Low-end systems: longer installation times

RAM usage during reconstruction: 2–8 GB

Expected loading time reduction: up to 50%

Section 4: Testing Scenarios

4.1 3D Model Compression Test

Select a model with 1,000,000 vertices

Encode with Quantization + Delta Encoding + RGB Channel

Reconstruct with ML and reload on GPU

Compare original vs. reconstructed geometry

4.2 Texture Reconstruction Test

Select a heavy texture (4K or 8K)

Convert to Base Texture + Recipt

Rebuild final texture during installation

Compare quality with the original (resolution, detail, color fidelity)

Section 5: Case Study – ARK: Survival Evolved
Asset Category Original Size (GB) After Compression (GB)
HD Textures 200 20
3D Models 50 5
Audio, Animation 100 100
Engine & Overhead 85 50
Total 435 175

Key Results:

Texture download size reduced: 200 → 20 GB

Model size reduced: 50 → 5 GB

Total reduction: 435 → 175 GB (≈60%)

Strategic Impacts:

~260 GB saved for players

Faster downloads, improved user trust

Optimized runtime performance, lower GPU/RAM load

Cost savings for publishers due to reduced bandwidth

Standards and Compression Formats

ASTC (Adaptive Scalable Texture Compression): flexible, high-quality, from mobile to PC

BC3 (Block Compression 3 / DXT5): widely supported, suitable for textures with alpha channels

Conclusion :

3D Model Reconstruction: up to 90% reduction, GPU-friendly

Texture Reconstruction: up to 90% download reduction, full-quality restoration with Recipt

Combined Advantage: overall game size reduction, improved user experience, easier customization and updates

Investment Value:
These methods are innovative, efficient, and promising for publishers and investors.