I’m building a feature that takes a photo of a bedroom and instantly tells the user whether it contains a bed, wardrobe or nightstand. To keep the mobile app lean, I need a very lightweight computer-vision model—something that can run quickly on-device without a large memory footprint, yet still remain reliable across common bedroom layouts, angles and lighting conditions. Here’s what matters most to me: • The model must accurately detect and label the three furniture classes: bed, wardrobe and nightstand. • It should work on single images (not video) and return bounding boxes or masks so I can highlight each item in the UI. • Smaller is better: please target a footprint that can comfortably fit into a typical smartphone package while keeping inference times snappy. • I’ll need the trained model file, the training notebook or script, and a short README that explains how to reproduce the training and run inference. If you already have experience with MobileNet, EfficientDet, YOLO-Nano, TensorFlow Lite or similar tiny-model workflows, your expertise will be valuable here. Accuracy is important, but compactness is equally critical, so let me know what trade-offs you recommend and past results you’ve achieved on similar lightweight object-detection tasks. When you reply, please outline: 1. Your preferred architecture and why it suits this job. 2. Expected final model size and typical inference speed on a mid-range phone or Raspberry Pi-class device. 3. Any data requirements or augmentation you’ll need from me. Once we agree on the approach, I’ll share a small curated dataset of bedroom images to get us started, and we can iterate until the detections are solid.