AN 998: Intel® FPGA AI Suite Inference on Quantized Graphs

ID 786296
Date 12/01/2023
Public

1.3. Preparing the imagenette/160px-V2 Dataset 

In this example, you run inference with quantized ResNet-18 on the imagenette/160px-V2 dataset, which is a small subset of Imagenet. For details, visit Imagenette at github.com.

The Intel® FPGA AI Suite provides a script to download and organize the validation subset. The script requires the COREDLA_WORK environment variable to be set. To use ~/coredla_work/ as your working directory, run the following commands:
mkdir ~/coredla_work
cd ~/coredla_work
source dla_init_local_directory.sh
To run the script, issue the following commands:
source /opt/intel/openvino_2022/setupvars.sh
source /opt/intel/fpga_ai_suite_2023.2/dla/bin/init_env.sh
/opt/intel/fpga_ai_suite_2023.2/dla/bin/download_imagenette.sh
When the script finishes, you should see the following message:
Imagenette dataset was downloaded successfully!
The final organized 3925 validation images should be located in the following directory:
 $COREDLA_WORK/demo/imagenette/imagenette2-160/organized/

This directory also contains the correct_classifications.txt file that contains the correct class labels for the images. The Intel® FPGA AI Suite uses this file to compute final accuracy.