This dataset consists of traffic light images collected from Map10HD of the CARLA simulator and is intended for training traffic light classification models. A simulated vehicle equipped with eight cameras was driven within the CARLA simulator, and YOLO11s was used to extract the images. Each image measures 64 pixels in height and 32 pixels in width, with a total of 18,201 images. There are five categories in total: 0 (red), 1 (yellow), 2 (green), 3 (unknown), and 4 (not_traffic_light). A "label.csv" file was generated through manual classification of all the images.