Fig 1.
Examples of body postures of frogs and toads on land (A-C) and in water (in natural environment: D-F, or in an unfamiliar white box: G-I).
While legs are typically tightly folded on land, body postures in water often provide better visibility of them. Species: agile frog (Rana dalmatina; A, D, G), common toad (Bufo bufo; B, H), marsh frog (Pelophylax ridibundus; C), Colombian four-eyed frog (Pleurodema brachyops; E), masked tree frog (Smilisca phaeota; F), and African clawed frog (Xenopus laevis; I). Photo credit: Nikolett Ujhegyi (A, B, H), Charles J. Sharp (C, F), Edina Nemesházi (D, G), Luis Alberto Rueda (E) and Zsanett Mikó (I).
Fig 2.
Agile frog individuals with various skin tones photographed after paper-towel drying (top) and under water (bottom).
For each individual, the two images were captured a few seconds apart, using the same light source, while avoiding direct-light exposure to minimise reflections. Note how submerging in water decreases skin glossiness and shadows caused by uneven body surface, overall improving the perception of contrast between skin tone and pigment patterns. For illustration purposes on this figure, we applied the same exposure curve across all images in RawTherapee (referred to as ‘baseline’ on S2 Fig), and cropped them subsequently.
Fig 3.
Fine-tuned photography of frogs submerged in water in 2024.
The two poses are shown in water medium: ‘free’ (A) and ‘restrained’ (B). The images were cropped to indicate the approximate coverage of the frog body that was used for ID by HotSpotter. Panel C shows visual representation of key methodological components for both image types: frogs were photographed in water, while direct light exposure was physically prevented.
Table 1.
False rejection rates by HotSpotter in different databases.
Table 2.
Results of the Bayesian generalized linear mixed-effects models with FRR values reflecting software performance, calculated either based on the original ranks for all databases (FRR), or based on corrected ranks in the ‘dry-restrained 2024’ but original ranks in all the other databases (Corrected FRR).
Fig 4.
Distribution of ‘hotspots’ displayed by HotSpotter.
Image pairs feature the same female after (on the let) and before (on the right) egg deposition in the ‘dry-restrained’ (A), ‘water-restrained’ (B) and ‘water-free’ (C) databases of females in 2024. The colour of the ovals denoting ‘hotspots’ indicate similarity strength scores increasing from dark red to bright yellow as shown on the colour bars under the images (A: 50-150, B: 80-240, C: 60-180; explanation taken from [6]). Although all ‘hotspots’ are located on the animal in each panel, those ovals in one image that have no matching oval at the same location in the other image denote false ‘hotspots’ (we marked these by asterisks).