Developing Your Models

Use the following general process guidelines to develop real-time video processing models to run on embedded targets. By optimizing the model at each step, you improve its final performance.

1. Create the initial model and optimize the implementation algorithm. Use floating-point data types so that the model runs faster in simulation mode. If you are working with a floating-point processor, go to step 3.

2. If you are working with a fixed-point processor, gradually change the model data types to fixed point, and run the model after every modification.

   During this process, you can use data type conversion blocks to isolate the floating point sections of the model from the fixed-point sections. You should see a performance improvement if you run the model in accelerator mode.

3. Remove unnecessary sink blocks, including scopes, and blocks that log data to files.

4. Compile the model for deployment on the embedded target.