Fast Single Classification but with width, height, x, y

[Rocksetta]

The multi-object detection is excellent as it gives x,y, width and height for the object, but in my situation it is much slower than the single object method. Is there anyway to get x, y, width and height for a single object? Presently the multi-object detection is for 320 x 320, can it be used on a smaller resolution?

Explanation: A lot of the work my students do is around robot cars and now robot arms and fast object detection like the Pixy2 achieves is very useful, but the Pixy2 just detects colours not objects. Fast object detection and location would be very useful. Any suggestions for speeding up the multi-detect.

Come to think of it you can probably set the multi-detect to only detect one object, that might speed it up.

[janjongboom]

Not really, the underlying network is just much more involved, and thus requires a lot more compute. We’ll be coming out with new and smaller transfer learning models in the near future which would fix that (and bring this also to Cortex-M boards!).

[Rocksetta]

So I tried putting the multi-object detection Impulse onto my Portenta Vision Shield and it was too big, so I switched to GRAYSCALE and got this error.

@janjongboom you mention working on smaller transfer learning models, I like the 320 x 320 size but is there a plan to also support the Arduino Portenta with some form of GRAYSCALE?

[janjongboom]

@Rocksetta, yes, we’ll come with grayscale models too when we update the object detection models for Cortex-M.