tensorflow-example-java Some silly questions

This looks really cool, how do I see the box around the item it detects?

How easy would it be to feed a webcam or video in to it and make it do it realtime?

Jan 19 '19 14:01 burf2000

@burf2000 Same question. Have you done it?

Mar 30 '19 02:03 dreistheman

Hi everybody, It should not be to difficult to modify the project to process a video or a webcam stream. You only need a library to read video or webcam streams and you have to execute the algorithm for each images. Something similar is implemented in the android version, where the application reads from the mobile phones camera stream and process it. Please have a look at the project here: https://github.com/szaza/android-yolo-v2. I would also mention that the darknet framework is able to run object detection on video and webcam streams: https://pjreddie.com/darknet/yolo/.

Apr 01 '19 06:04 szaza

@szaza As you've said, it could be done straightforwardly with a library for video iteration like OpenCV's VideoCapture class. Only concern now is that the inference is very slow like 12 seconds per image frame. I'm working on a surveillance (desktop) app that intends to use YOLO Java for object detection with further subsequent analysis. (OpenCV, Java). I'm wondering how your Android-Yolo-v2 can run in real-time, is it the TensorflowInferenceInterface?

May 03 '19 03:05 dreistheman