iDetection uses your iOS device wide-angle camera, and applies the latest realtime AI Object Detection algorithm to the scene to detect and locate up to 80 classes of common objects. All processing is done directly on the iOS device, no cloud computation is performed. Your images are not transmitted off your device.
The AI object detector we use is a deep neural network called YOLOv3-SPP (You Only Look Once v3 with Spatial Pyramid Pooling). Originally designed by Joseph Redmon, YOLOv3-SPP is trained in PyTorch and transferred to an Apple CoreML model via ONNX.
A12 iOS device performance is up to 30 FPS at the default 192 x 320 pixel image size. Performance may be significantly slower on older devices.
For more info please visit: https://github.com/ultralytics/yolov3
-Bug fixes and performance improvements
Ratings and Reviews
Hi really good work.
But I want to try change image resolution and to see the performance. How does it happen?
Developer Response ,
Thank you!! We are now packaging the app with both YOLOv3-SPP and YOLOv3-tiny. Tiny should feature faster FPS.
Actual changes to the inference shape at runtime should theoretically be possible using CoreML flexible input sizes, though we haven't succeeded in this so far. Help is welcome if you are an iOS developer!
This app works great! The FPS raises and lowers as more objects come into view, but they fixed that with max items slider with 1 objects detected at a time to 100! So great job developers!
It’s like a bit of magic on my iPhone XR.
With Family Sharing set up, up to six family members can use this app.