[APAI23-LAB09] End-to-end CNN deployment

1. Setup

Download the repo

Open the usual VM used for the other labs, then:

$ cd <work_dir>
$ git clone https://github.com/EEESlab/APAI24-Old-LAB09-end-to-end-CNN-deployment.git --recurse-submodules --shallow-submodules

Note: make sure to clone with --recurse-submodules !

Setup PULP-SDK and related

Source:

$ source setup_pulp_sdk.sh

Install dependencies

$ pip install onnx ortools 

2. How to run the code

1. cd to dory directory

$ cd APAI24-Old-LAB09-end-to-end-CNN-deployment/dory

2. Generate the network

$ python network_generate.py Quantlab <target> ../<network>/config.json --perf_layer

The target can be:

• PULP.PULP_gvsoc - 8-core cluster execution
• PULP.PULP_NE16 - NE16 execution

The network can be:

• Network1 - network with a 3x3 convolution for the last layer

• Network2 - network with a 5x5 convolution for the last layer

3. Build and run the code

$ make -C application clean all run CORE=8