Any network suit for detect thousands object on mobile phone?

As the title mentioned,any network suit for this kind of task? Check yolo 9000, looks like it need more than 3GB of ram to perform the inference task, do you have any recommendation for this kind of task?Thanks

ps : The last layer of the yolo 9000 shown on this page only cost 110.43Mbytes(28947456/1024/1024*4), not sure why it could consume more than 3GB when doing inference.

Make sure you’re correctly resizing the spatial dimensions of the input to the network.

Check out the yolo3_mobilenet1.0_coco 3 (320x320) pretrained model from GluonCV for example. It should take ~50Mb for a batch of 128 images of resolution 320x320.

Thanks, test with cpu version, it use less than 200MBytes of ram, maybe 3GB of ram is for training but not inference.