OlafenwaMoses/ImageAI — reverse-engineered prompt

Build me a simple Python computer vision library that makes common AI image tasks feel easy, even for someone who is not deep into machine learning. I want to be able to load a pretrained model and do image classification, object detection on photos, and object detection on videos with just a few lines of code. It should work offline, support normal CPU use, and leave room for GPU use too.

Please include the main image prediction options shown in the project, like MobileNetV2, ResNet50, InceptionV3, and DenseNet121, plus object detection with RetinaNet, YOLOv3, and TinyYOLOv3 where appropriate. I also want support for custom model training for classification and detection, or at least a clean starting point for it.

Make it feel like a real open source package, with clear examples, sample images or videos, sensible defaults, and basic tests. If there are older backend pieces, keep them separate from the current version. Look up current docs online if you need to.