I have a Mobile SSD algorithm to do people detection and framing, the model size is 22mb, I want to compress it as small as possible without losing too much accuracy. You can use methods like channel-pruning, changing the model, and transform the float number to int.
What I need: source code to modify the large model, not only for this project.
The introduction of the algorithm:
I placed a camera on top of my TV covering my whole living room. It can detect people (one or multiple) in the room, frame them, and count the number in real time.
We use Mobile net + SSD to realize this.