04_texture_compression_app

04 — Texture Compression App

Real-time windowed application that trains an MLP to compress a 2D texture on the GPU. Training runs incrementally (a configurable number of epochs per frame), and the reconstructed texture is displayed live as a fullscreen quad with an ImGui overlay for monitoring and parameter control.

Based on Example 03 but restructured for interactive use.

Building

cmake --build build --target 04-texture-compression-app --config Release

Requires GFX_ENABLE_GUI ON in third_party/gfx_dep/CMakeLists.txt for ImGui support.

Usage

04-texture-compression-app [options]

Examples

# Default: train on a checkerboard pattern
04-texture-compression-app

# Train on a PNG image
04-texture-compression-app --input-image photo.png

# Faster training: 5 epochs per frame with larger batch
04-texture-compression-app --input-image photo.png --epochs-per-frame 5 --batch-size 50000

# Positional encoding with more frequency bands
04-texture-compression-app --input-encoding positional --positional-frequencies 8

# Identity (raw UV) encoding
04-texture-compression-app --input-encoding none

Command-Line Options

Option	Default	Description
--backbone-layers	4	Number of MLP backbone layers
--hidden-dim	64	Hidden layer dimension
--activation	leaky_relu	Activation: identity, sigmoid, tanh, relu, leaky_relu
--bias / --no-bias	true	Enable/disable bias in MLP layers
--seed	987654321	Random seed
--samples	200000	Number of training samples
--batch-size	20000	Training batch size (adjustable at runtime via ImGui)
--learning-rate	0.005	Optimizer learning rate
--optimizer	adam	Optimizer: sgd, adam, lion
--input-encoding	positional	Input encoding: none, positional
--positional-frequencies	4	Frequency bands for positional encoding
--input-image	—	PNG image to use as ground truth
--texture-width	2048	Generated texture width (ignored with --input-image)
--texture-height	2048	Generated texture height (ignored with --input-image)
--texture-pattern	checkerboard	Pattern: gradient, checkerboard, stripes, circle, perlin
--epochs-per-frame	1	Training epochs per frame
--window-width	1024	Window width
--window-height	1024	Window height
--loss-scale	512	Loss scale for FP16 gradient stability
--software-linalg	false	Use software linear algebra on HLSL
--debug	false	Enable GPU debug/shader debugging

ImGui Controls

• Start/Pause Training — training starts paused; click to begin
• Encoding — switch between None (identity) and Positional encoding (requires reinit)
• Frequencies — positional encoding frequency count (shown when Positional is selected)
• Batch Size — adjustable at any time, takes effect immediately
• Loss graph — log-scale loss plotted over epochs
• Stats — current epoch, loss, training time, inference time, FPS

Key Differences from Example 03

• Windowed with live display instead of headless
• GPU-only (no C++ fallback path)
• Incremental training across frames
• Interactive parameter tuning via ImGui
• No output PNG — display is live on screen