Add mlperf_tiny benchmark targets

Change-Id: I38505ccc64da6168804ac56d8582c9791069f53a
2 files changed
tree: 4db6495fa4ebe7fa2df5e20e5d82e928513b75b7
  1. benchmarks/
  2. float_models/
  3. quant_models/
  4. test_data/
  5. .bazelrc
  6. .bazelversion
  7. .gitignore
  8. BUILD
  9. CMakeLists.txt
  10. CONTRIBUTING.md
  11. LICENSE
  12. opentitan_sanitized_requirements.txt
  13. PREUPLOAD.cfg
  14. README.md
  15. WORKSPACE
README.md

Public ML model zoo

This is the model zoo for public models used in Shodan vector core examples.

Model discription

person_detection.tflite

The person presence detection quantized model from https://github.com/tensorflow/tflite-micro/blob/main/tensorflow/lite/micro/models/person_detect.tflite

mobilenet_v1_0.25_224_float.tflite

Mobilenet V1 float-point model from https://tfhub.dev/tensorflow/lite-model/mobilenet_v1_0.25_224/1/default/1?lite-format=tflite

mobilenet_v1_0.25_224_quant.tflite

Mobilenet V1 quantized model from https://tfhub.dev/tensorflow/lite-model/mobilenet_v1_0.25_224_quantized/1/default/1?lite-format=tflite

mobilenet_v2_1.0_224_quant.tflite

Mobilenet V2 quantized model from https://tfhub.dev/tensorflow/lite-model/mobilenet_v2_1.0_224_quantized/1/default/1?lite-format=tflite

hps_quant.tflite

HPS (Human Presence Sensor) non-tiled quantized model from https://chromium.googlesource.com/chromiumos/platform/hps-firmware/+/6cdea6d1158a8cd3238b8ae4f744fdb494779c80/models/shared.tflite

Visualize the model

For tflite models, use the web-based visualization tool to inspect the file. The tool supports drag and drop or file GUI.

Test data

Test data for sample model inputs are in the “test_data” directory. Currently, only test data for the Human Presence Sensor model are included.

hps_0-6.jpg

Test image files for Human Presence Sensor model, from ChromeOS codebase.

Model executables

For each model, the corresponding model executable under iree_exec can be built with the .c/.h files using sw/vec_iree project as library. They can be run as unit tests via lit framework using Renode or QEMU.