Image Classification Code Sample


This document describes the process of setting up and running the Arm® Ethos™-U55 Image Classification example.

This use-case example solves the classical computer vision problem of image classification. The ML sample was developed using the MobileNet v2 model that was trained on the ImageNet dataset.

Use-case code could be found in the following directory:source/use_case/img_class.


See Prerequisites

Building the code sample application from sources

Build options

In addition to the already specified build option in the main documentation, the Image Classification use-case specifies:

  • img_class_MODEL_TFLITE_PATH - The path to the NN model file in the TFLite format. The model is then processed and included in the application axf file. The default value points to one of the delivered set of models.

    Note that the parameters img_class_LABELS_TXT_FILE,TARGET_PLATFORM, and ETHOS_U55_ENABLED must be aligned with the chosen model. In other words:

    • If ETHOS_U55_ENABLED is set to On or 1, then the NN model is assumed to be optimized. The model naturally falls back to the Arm® Cortex®-M CPU if an unoptimized model is supplied.
    • if ETHOS_U55_ENABLED is set to Off or 0, the NN model is assumed to be unoptimized. Supplying an optimized model in this case results in a runtime error.
  • img_class_FILE_PATH: The path to the directory containing the images, or a path to a single image file, that is to be used in the application. The default value points to the resources/img_class/samples folder containing the delivered set of images.

    For further information, please refer to: Add custom input data section.

  • img_class_IMAGE_SIZE: The NN model requires input images to be of a specific size. This parameter defines the size of the image side in pixels. Images are considered squared. The default value is 224, which is what the supplied MobilenetV2-1.0 model expects.

  • img_class_LABELS_TXT_FILE: The path to the text file for the label. The file is used to map a classified class index to the text label. The default value points to the delivered labels.txt file inside the delivery package. Change this parameter to point to the custom labels file to map custom NN model output correctly.

  • img_class_ACTIVATION_BUF_SZ: The intermediate, or activation, buffer size reserved for the NN model. By default, it is set to 2MiB and is enough for most models.

  • USE_CASE_BUILD: is set to img_class to only build this example.

To build ONLY the Image Classification example application, add -DUSE_CASE_BUILD=img_class to the cmake command line, as specified in: Building.

Build process

Note: This section describes the process for configuring the build for the MPS3: SSE-300. To build for a different target platform, please refer to: Building.

Create a build directory and navigate inside, like so:

mkdir build_img_class && cd build_img_class

On Linux, when providing only the mandatory arguments for the CMake configuration, execute the following command to build only Image Classification application to run on the Ethos-U55 Fast Model:

cmake ../ -DUSE_CASE_BUILD=img_class

To configure a build that can be debugged using Arm DS specify the build type as Debug and then use the Arm Compiler toolchain file:

cmake .. \
    -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/toolchains/bare-metal-armclang.cmake \

For further information, please refer to:

Note: If re-building with changed parameters values, we recommend that you clean the build directory and re-run the CMake command.

If the CMake command succeeds, build the application as follows:

make -j4

To see compilation and link details, add VERBOSE=1.

Results of the build are placed under the build/bin folder, like so:

 ├── ethos-u-img_class.axf
 ├── ethos-u-img_class.htm
 └── sectors
      ├── images.txt
      └── img_class
           ├── ddr.bin
           └── itcm.bin

The bin folder contains the following files:

  • ethos-u-img_class.axf: The built application binary for the Image Classification use-case.

  • Information from building the application. For example: The libraries used, what was optimized, and the location of objects.

  • ethos-u-img_class.htm: Human readable file containing the call graph of application functions.

  • sectors/img_class: Folder containing the built application. It is split into files for loading into different FPGA memory regions.

  • sectors/images.txt: Tells the FPGA which memory regions to use for loading the binaries in the sectors/.. folder.

Add custom input

The application anomaly detection is set up to perform inferences on data found in the folder, or an individual file, that is pointed to by the parameter img_class_FILE_PATH.

To run the application with your own images, first create a folder to hold them and then copy the custom images into the following folder:

mkdir /tmp/custom_images

cp custom_image1.bmp /tmp/custom_images/

Note: Clean the build directory before re-running the CMake command.

Next, set img_class_FILE_PATH to the location of this folder when building:

cmake .. \
    -Dimg_class_FILE_PATH=/tmp/custom_images/ \

The images found in the img_class_FILE_PATH folder are picked up and automatically converted to C++ files during the CMake configuration stage. They are then compiled into the application during the build phase for performing inference with.

The log from the configuration stage tells you what image directory path has been used:

-- User option img_class_FILE_PATH is set to /tmp/custom_images
-- User option img_class_IMAGE_SIZE is set to 224
-- Generating image files from /tmp/custom_images
++ Converting custom_image1.bmp to
-- Defined build user options:
-- img_class_FILE_PATH=/tmp/custom_images
-- img_class_IMAGE_SIZE=224

After compiling, your custom images have now replaced the default ones in the application.

Note: The CMake parameter IMAGE_SIZE must match the model input size. When building the application, if the size of any image does not match IMAGE_SIZE, then it is rescaled and padded so that it does.

Add custom model

The application performs inference using the model pointed to by the CMake parameter MODEL_TFLITE_PATH.

Note: If you want to run the model using an Ethos-U55, ensure that your custom model has been successfully run through the Vela compiler before continuing.

For further information: Optimize model with Vela compiler.

To run the application with a custom model, you must provide a labels_<model_name>.txt file of labels that are associated with the model. Each line of the file must correspond to one of the outputs in your model.

Refer to the provided labels_mobilenet_v2_1.0_224.txt file for an example.

Then, you must set img_class_MODEL_TFLITE_PATH to the location of the Vela processed model file and img_class_LABELS_TXT_FILE to the location of the associated labels file.

For example:

cmake .. \
    -Dimg_class_MODEL_TFLITE_PATH=<path/to/custom_model_after_vela.tflite> \
    -Dimg_class_LABELS_TXT_FILE=<path/to/labels_custom_model.txt> \

Note: Clean the build directory before re-running the CMake command.

The .tflite model file pointed to by img_class_MODEL_TFLITE_PATH, and the labels text file pointed to by img_class_LABELS_TXT_FILE are converted to C++ files during the CMake configuration stage. They are then compiled into the application for performing inference with.

The log from the configuration stage tells you what model path and labels file have been used, for example:

-- User option img_class_MODEL_TFLITE_PATH is set to <path/to/custom_model_after_vela.tflite>
-- User option img_class_LABELS_TXT_FILE is set to <path/to/labels_custom_model.txt>
-- Using <path/to/custom_model_after_vela.tflite>
++ Converting custom_model_after_vela.tflite to\
-- Generating labels file from <path/to/labels_custom_model.txt>
-- writing to <path/to/build/generated/src/>

After compiling, your custom model has now replaced the default one in the application.

Setting up and running Ethos-U55 code sample

Setting up the Ethos-U55 Fast Model

The FVP is available publicly from Arm Ecosystem FVP downloads.

For the Ethos-U55 evaluation, please download the MPS3 version of the Arm® Corstone™-300 model that contains both the Ethos-U55 and Cortex-M55. The model is currently only supported on Linux-based machines.

To install the FVP:

  • Unpack the archive.

  • Run the install script in the extracted package:

  • Follow the instructions to install the FVP to the required location.

Starting Fast Model simulation

The pre-built application binary ethos-u-img_class.axf can be found in the bin/mps3-sse-300 folder of the delivery package.

Assuming that the install location of the FVP was set to ~/FVP_install_location, then the simulation can be started by using:


A log output appears on the terminal:

telnetterminal0: Listening for serial connection on port 5000
telnetterminal1: Listening for serial connection on port 5001
telnetterminal2: Listening for serial connection on port 5002
telnetterminal5: Listening for serial connection on port 5003

This also launches a telnet window with the standard output of the sample application. It also includes error log entries containing information about the pre-built application version, TensorFlow Lite Micro library version used, and data types. The log also includes the input and output tensor sizes of the model compiled into the executable binary.

After the application has started, if img_class_FILE_PATH points to a single file, or even a folder that contains a single image, then the inference starts immediately. If there are multiple inputs, it outputs a menu and then waits for input from the user:

User input required
Enter option number from:

1. Classify next image
2. Classify image at chosen index
3. Run classification on all images
4. Show NN model info
5. List images


What the preceding choices do:

  1. Classify next image: Runs a single inference on the next in line image from the collection of the compiled images.

  2. Classify image at chosen index: Runs inference on the chosen image.

    Note: Please make sure to select image index from within the range of supplied audio clips during application build. By default, a pre-built application has four images, with indexes from 0 to 3.

  3. Run classification on all images: Triggers sequential inference executions on all built-in images.

  4. Show NN model info: Prints information about the model data type, input, and output, tensor sizes:

    INFO - uTFL version: 2.5.0
    INFO - Model info:
    INFO - Model INPUT tensors:
    INFO -  tensor type is UINT8
    INFO -  tensor occupies 150528 bytes with dimensions
    INFO -    0:   1
    INFO -    1: 224
    INFO -    2: 224
    INFO -    3:   3
    INFO - Quant dimension: 0
    INFO - Scale[0] = 0.007812
    INFO - ZeroPoint[0] = 128
    INFO - Model OUTPUT tensors:
    INFO -  tensor type is UINT8
    INFO -  tensor occupies 1001 bytes with dimensions
    INFO -    0:   1
    INFO -    1: 1001
    INFO - Quant dimension: 0
    INFO - Scale[0] = 0.098893
    INFO - ZeroPoint[0] = 58
    INFO - Activation buffer (a.k.a tensor arena) size used: 521760
    INFO - Number of operators: 1
    INFO -  Operator 0: ethos-u
  5. List Images: Prints a list of pair image indexes. The original filenames are embedded in the application, like so:

    INFO - List of Files:
    INFO - 0 => cat.bmp
    INFO - 1 => dog.bmp
    INFO - 2 => kimono.bmp
    INFO - 3 => tiger.bmp

Running Image Classification

Please select the first menu option to execute Image Classification.

The following example illustrates an application output for classification:

INFO - Running inference on image 0 => cat.bmp
INFO - Final results:
INFO - Total number of inferences: 1
INFO - 0) 282 (14.636096) -> tabby, tabby cat
INFO - 1) 286 (14.537203) -> Egyptian cat
INFO - 2) 283 (12.757138) -> tiger cat
INFO - 3) 458 (7.021370) -> bow tie, bow-tie, bowtie
INFO - 4) 288 (7.021370) -> lynx, catamount
INFO - Profile for Inference:
INFO - NPU ACTIVE cycles: 7489258
INFO - NPU IDLE cycles: 914
INFO - NPU TOTAL cycles: 7490172

It can take several minutes to complete one inference run. The average time is around 2-3 minutes.

The log shows the inference results for image 0, so 0 - index, that corresponds to cat.bmp in the sample image resource folder.

The profiling section of the log shows that for this inference:

  • Ethos-U55 PMU report:

    • 7,490,172 total cycle: The number of NPU cycles.

    • 7,489,258 active cycles: The number of NPU cycles that were used for computation.

    • 914 idle cycles: The number of cycles for which the NPU was idle.

    • 2,489,726 AXI0 read beats: The number of AXI beats with read transactions from AXI0 bus. AXI0 is the bus where the Ethos-U55 NPU reads and writes to the computation buffers, activation buf, or tensor arenas.

    • 1,098,726 AXI0 write beats: The number of AXI beats with write transactions to AXI0 bus.

    • 471,129 AXI1 read beats: The number of AXI beats with read transactions from AXI1 bus. AXI1 is the bus where the Ethos-U55 NPU reads the model. So, read-only.

  • For FPGA platforms, a CPU cycle count can also be enabled. However, do not use cycle counters for FVP, as the CPU model is not cycle-approximate or cycle-accurate

The application prints the top five classes with indexes, a confidence score, and labels from the associated labels_mobilenet_v2_1.0_224.txt file. The FVP window also shows the output on its LCD section.