(1105) Principles of Beautiful Figures for Research Papers - YouTube

1. Quality and Vector Graphics

The author stresses using vector-based images (e.g., PDF format) over raster ones (e.g., JPEG or PNG) to maintain quality, as raster images can pixelate or compress during publication. He shares a personal anecdote from a 2019 conference where his figures degraded post-submission, appearing pixelated in the final version despite looking fine initially. Vector graphics, defined by shapes and formulas, remain scalable without quality loss, allowing readers to zoom in seamlessly.

• Recommendations: Always check final proofs for compression issues. For complex data (e.g., 15,000 points), combine vector elements with raster for efficiency, as seen in his 2024 paper.
• Examples: He demonstrates zooming into his vector figures from recent manuscripts, showing no pixels, versus degraded raster ones.

2. Readability

Figures must be legible without excessive zooming, using fonts no smaller than the journal's smallest text (e.g., 8 points). The author advises checking by comparing font sizes with a cursor or viewing from a distance.

• Examples of Issues: In a 2024 IEEE Access paper, tiny fonts on maps and subfigures required 300% zoom to read, violating readability. Another paper's heat map needed 1,200% zoom to discern details.
• Good Practices: His own figures match manuscript font sizes, ensuring text like axis labels is clear at 100% view.

3. Simplify and Declutter

Overly complex figures confuse readers, so remove unnecessary elements and focus on hierarchy. Ask: Is this essential? Can it be simplified or moved to an appendix? Get feedback to ensure clarity.

• Examples of Clutter: A 2024 framework diagram with excessive blocks, arrows, and images lacked structure. "Spaghetti plots" with too many overlapping lines were meaningless.
• Better Approaches: The author simplified his frameworks to 4-5 blocks with clear flow (e.g., input-modeling-output). In an OpenAI GPT-3 paper, authors faded background lines and highlighted key ones for focus. Use numbers or loops to guide readers through complex processes.

4. Colors

Colors should enhance, not overwhelm. Design in black, white, and gray first, adding 1-2 colors sparingly. Avoid bright, complementary colors (e.g., red and green) that clash; prefer analogous palettes for harmony.

• Tools and Tips: Use Adobe Color Wheel for harmonies or ColorHexa for palettes. In Julia or Python, select built-in gradients like blue-to-dark-blue, avoiding rainbows. Consider color psychology: red signals "bad," green "good"; yellow fades on white backgrounds → don’t use yellow color.
• Examples: A 2024 paper's rainbow heat maps confused viewers. Churkin's gray-to-red gradients intuitively showed declining areas as "dangerous". A 2007 wind forecast paper used only shades of gray effectively.

5. Message and Story

Figures aren't just data dumps; they should tell a story. Strategically use colors, text, and layout to guide readers, often in a "Z-shape" reading pattern (top-left to bottom-right). Highlight key insights in descriptions and labels.

• Examples: A Jacobian error figure lacked axis labels and guidance, leaving readers lost. In contrast, a molecular contribution heat map used a color bar upfront to explain the narrative intuitively.