Rodent Behavior Classification Dataset

This dataset contains approximately 116,000 RGB images (128×128 px) extracted from video recordings of seven Roman High- and Low-Avoidance rats observed individually in Skinner boxes (operant conditioning chambers). Each chamber includes a lever–hole module and a feeding window, allowing the observation of spontaneous behaviors under controlled conditions. The dataset has been designed to support the development and evaluation of computer vision models for automated behavior recognition in rodents. Each frame has been labeled into one of seven behavioral classes: Floor-Explore, Floor-Poke, Floor-Bite, Stand-Explore, Stand-Poke, Stand-Bite, and Stand-Eat. These classes represent combinations of two postures (Floor or Stand) and specific actions: exploratory movement, interaction with the lever hole, biting behavior, and food-related activity at the feeding window. All recordings were collected following institutional animal welfare and ethical standards. The dataset provides a compact, well-curated benchmark for image-based behavioral classification tasks in laboratory rodents.