Reproduce “Combining Semantic Guidance and Deep Reinforcement Learning For Generating Human Level Paintings” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jaskirat Singh, Liang Zheng
Researchers at the Australian National University developed a deep reinforcement learning framework incorporating semantic guidance to generate stroke-based paintings that capture fine-grain features and handle diverse object variations. The approach yields canvases with higher semantic segmentation accuracy and a human-like "top-down" painting process, outperforming baselines on complex datasets.
Key insights
- Semantic guidance, through bi-level painting and neural alignment, significantly improves the agent's ability to differentiate and focus on specific semantic regions.
- A focus reward derived from guided backpropagation effectively enhances the depiction of fine-grain features in the generated artwork.
- The semantically guided agent adopts a "top-down" painting strategy, starting with broad structures and refining details, mirroring human artistic processes more closely than "bottom-up" baselines.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.