Lab rat thermal imaging dataset enables automated anatomical segmentation
1,655 radiometric frames with pixel-level head, body, and tail masks—plus a U-Net baseline at 0.895 IoU.
A team of researchers has released a radiometric thermal imaging dataset of 25 laboratory rats, containing 1,655 quality-controlled frames. Each 480×640 frame pairs raw temperature matrices in Celsius with dense four-class anatomical masks (background, head, body, and tail). Drawn from ethanol and ketamine cohorts—interventions that alter thermoregulation in opposite directions—the dataset captures a wide range of surface temperature regimes, with per-class temperatures following a head > body > tail ordering. An exploratory U-Net segmentation pipeline achieves a subject-level cross-validated mean intersection-over-union of 0.895 ± 0.006, making the dataset a reuse-ready benchmark for thermal semantic segmentation and downstream physiological and stress-phenotyping analyses.
- 1,655 radiometric thermal frames from 25 lab rats with 480×640 raw temperature matrices in Celsius
- Four-class pixel-level segmentation masks (background, head, body, tail) aligned to physical temperatures
- Baseline U-Net achieves 0.895 mIoU; data spans ethanol and ketamine cohorts for diverse thermoregulatory states
Why It Matters
Automates lab-animal thermal analysis, accelerating pharmacological and stress research with a reusable public benchmark.