Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Content Masked Loss: Human-Like Brush Stroke Planning in a Reinforcement Learning Painting Agent
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add Content Masking to clipped L1 loss (ablation) | Reinforcement Learning Painter · CelebA portraits, 30–200 strokes | human face-like judgments | Helped |
| Add Content Masking to L2 loss (ablation) | Reinforcement Learning Painter · CelebA portraits, 30–200 strokes | human face-like judgments | No clear effect |
| Add Content Masking to the loss (ablation) | Reinforcement Learning Painter · CelebA portraits, 50–200 strokes | FaceNet face detection | Helped |
| Use Content Loss on VGG-16 features instead of a pixel-based loss (ablation) | Reinforcement Learning Painter · Portrait painting | training outcome | Hurt |
| Use later VGG-16 layers for the Content Mask (ablation) | Content Masked Painter · Painting stroke sequence | human-like planning | Helped |
| Vary the VGG-16 layer used to create the Content Mask (ablation) | Content Masked Painter · Final painting | final painting results | No clear effect |
| Use Content Masking with clipped L1 rather than clipped L1 alone (ablation) | Reinforcement Learning Painter · CelebA held-out (2000 images) | FID: 241.58 -> 243.13 | Hurt |
| Use Content Masking with clipped L1 rather than clipped L1 alone (ablation) | Reinforcement Learning Painter · CelebA held-out (2000 images) | IS: 3.45 -> 3.47 | Helped |
| Use Content Masking with L2 rather than L2 alone (ablation) | Reinforcement Learning Painter · CelebA held-out (2000 images) | FID: 241.02 -> 242.67 | Hurt |
| Use Content Masking with L2 rather than L2 alone (ablation) | Reinforcement Learning Painter · CelebA held-out (2000 images) | IS: 3.05 -> 3.01 | Hurt |
Closest papers by meaning
- PrefPaint: Aligning Image Inpainting Diffusion Model with Human Preference (NeurIPS 2024 · similarity 0.84)
- Adversarial Perturbations Cannot Reliably Protect Artists From Generative AI (ICLR 2025 · similarity 0.84)
- Reinforcement Learning Platform for Adversarial Black-box Attacks with Custom Distortion Filters (AAAI 2025 · similarity 0.83)
- MaskSense: Confronting the Visual Exploration Trap in Masked Image Generation (NeurIPS 2026 · similarity 0.83)
- Goal Conditioned Reinforcement Learning for Photo Finishing Tuning (NeurIPS 2024 · similarity 0.83)
- Latent Representation Matters: Human-like Sketches in One-shot Drawing Tasks (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.