Skip to content

CHANGELOG.md

Latest commit

 

History

History
175 lines (121 loc) · 4.31 KB

CHANGELOG.md

File metadata and controls

175 lines (121 loc) · 4.31 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Added

Missing Paper Implementations:

  • Proper FVD (Fréchet Video Distance) implementation with I3D model (src/utils/fvd.py)
  • Human evaluation framework following paper's methodology (src/utils/human_eval.py)
  • Multi-scenario training support for diverse DOOM maps (src/environment/multi_scenario.py)

Developer Tools (scripts/):

  • download_models.py - Pre-download and verify all required models
  • resume_training.py - Easy resume from checkpoints
  • visualize_data.py - Visualize recorded gameplay data
  • compare_models.py - Compare different checkpoint quality
  • export_video.py - Batch export gameplay videos
  • monitor_training.py - Real-time training monitoring

Advanced Features (Beyond Paper):

  • Text-conditioned game generation with CLIP (src/diffusion/text_conditioning.py)
  • Image-based modding system for editing games (src/diffusion/image_modding.py)
  • Hierarchical memory system for longer context (src/diffusion/hierarchical_memory.py)

GitHub Infrastructure:

  • CONTRIBUTING.md - Contribution guidelines
  • GitHub Actions CI - Automated testing
  • Issue templates (bug report, feature request)
  • CHANGELOG.md - This file

Changed

  • Reorganized repository into standard GitHub structure
  • Moved configs to configs/ directory with clearer names
  • Moved tests to tests/ directory
  • Moved paper PDF to paper/ directory
  • Updated all import paths and references
  • Cleaned root directory to 6 essential files

[1.0.0] - 2025-10-27

Initial Release - Complete Implementation

Core Implementation

  • Action-conditioned Stable Diffusion model (943M parameters)
  • Training pipeline with noise augmentation
  • Real-time inference (4-step DDIM, 20 FPS)
  • PyTorch dataset for gameplay trajectories

Tier 1 - Chrome Dino

  • DQN agent implementation
  • SimpleDinoEnv wrapper
  • Training scripts
  • Complete in 2-3 days

Tier 2 - DOOM Lite

  • ViZDoom environment wrapper
  • PPO agent training
  • DOOM reward function
  • Complete in ~1 week

Tier 3 - Full DOOM

  • Adafactor optimizer (paper's choice)
  • Model distillation (1-step, 50 FPS)
  • Decoder fine-tuning
  • Comprehensive evaluation suite
  • Complete in 3-4 weeks

Documentation

  • Professional README
  • 12 comprehensive guides
  • Command references
  • Installation instructions
  • 5,000+ lines of documentation

Testing

  • Core component tests
  • All tier verification
  • CUDA/GPU validation
  • Installation tests

Configuration

  • 3 tier-specific configs
  • All hyperparameters documented
  • Easy customization

Technical Features

  • Noise augmentation for auto-regressive stability
  • Velocity parameterization training
  • Mixed precision (FP16) support
  • Classifier-Free Guidance
  • Auto-checkpointing and resume
  • TensorBoard logging
  • Comprehensive evaluation metrics (PSNR, LPIPS, SSIM)
  • Interactive gameplay mode
  • Video recording

Release Notes

v1.0.0 - Complete Implementation

First public release of complete GameNGen implementation.

Implements: "Diffusion Models Are Real-Time Game Engines" (ICLR 2025)

Status:

  • Implementation: Complete
  • Tests: All passing
  • Documentation: Comprehensive
  • Pretrained weights: Training in progress

What's Included:

  • All 3 tiers fully implemented
  • 12,000+ lines of production code
  • Ready to train immediately

Coming Soon:

  • Tier 1 weights (~3 days)
  • Tier 2 weights (~1 week)
  • Tier 3 weights (~4 weeks)
  • Demo videos

Future Roadmap

Planned Features

  • Pre-trained model weights for all tiers
  • Demo videos and comparisons
  • Jupyter notebook tutorials
  • Web-based demo interface
  • Additional game environments
  • Multi-game universal model
  • Longer context methods
  • Docker support

Research Extensions

  • Text-to-game generation
  • Image-based style transfer
  • Real-world applications (robotics, driving)
  • Improved architectures

Contributing

See CONTRIBUTING.md for guidelines on how to contribute.

Questions?

Open an issue or check the discussions.