Captions that stand alone

Many readers look at a figure before reading the text, or instead of it. The caption must let them understand the figure without hunting through the paper.

What a good caption has

1. A title sentence that states the finding, not just the topic. "Augmentation helps most when data is scarce." beats "Accuracy results."
2. What is shown: axes, groups, colours, symbols.
3. How much data: n per group, number of runs or seeds.
4. What error bars mean: SD, SE or 95% CI.
5. Any statistics shown (tests, significance marks).

An example

Example"Figure 2. Colour augmentation helps most when training data is scarce. Test accuracy (y-axis) against number of training images (x-axis, log scale) for models trained with (blue) and without (orange) colour augmentation. Points show the mean of 5 seeds; error bars show 95% confidence intervals. The gap closes above about 2,000 images."

Common mistakes

• A caption that only names the figure ("Results of experiment 2").
• Abbreviations defined only in the text.
• Error bars with no explanation.
• Saying "see text for details".

Tables need this too

Table captions (usually above the table) should say what the numbers are, their units, and what bold or stars mean.

RememberA reader should understand the figure from the figure and caption alone: finding first, then what's shown, n, and error bars.