Common scripts for benchmarking programs

Common preprocessing and postprocessing scripts to be used in benchmarking programs such as ck-tensorflow:program:object-detection-tflite, etc.

A client program has to reference scripts in its meta in the section run_time, e.g.:

  "run_cmds": {
    "default": {
      "run_time": {
        "pre_process_via_ck": {
          "data_uoa":       "24c98b0cee248d93",
          "module_uoa":     "script",
          "script_name":    "preprocess"
        },
        "post_process_via_ck": {
          "data_uoa":       "24c98b0cee248d93",
          "module_uoa":     "script",
          "script_name":    "postprocess"
        },

        "run_cmd_main": "$#BIN_FILE#$",

It is supposed that the client program provides required enviroment variables and dependencies with suitable names that scripts will search for.

Preprocessing

The preprocessing script prepares images for a client program.

Preprocessing steps:

• Read the required number of images from a dataset. The number of images is governed by program parameters CK_BATCH_COUNT and CK_BATCH_SIZE.

• Prepare images for loading into a model. Preparation includes scaling images to a size defined by the input images size of the model being benchmarked. See the section Input preprocessing parameters below.

• Store prepared images into a cache directory.

As a result, the preprocessing script provides a set of variables (in separate env.ini file) that a client program should use:

• MODEL_DATASET_TYPE - point to dataset type which the model was trained on (coco for example)
• MODEL_TFLITE_GRAPH - path to model's .tflite file
• MODEL_IMAGE_CHANNELS - number of color channels in image. Default value: 3
• MODEL_IMAGE_HEIGHT, MODEL_IMAGE_WIDTH - the sizes of an image, which it should have to be processed on by the model
• MODEL_NEED_BACKGROUND_CORRECTION - indicate that model don't consider "background" class (True or False)
• MODEL_NORMALIZE_DATA - indicate that image should be "normalized" before processing (True or False)

• MODEL_SUBTRACT_MEAN - indicate that operation "subtract mean" should be executed on the image before processing

• DETECTIONS_OUT_DIR - path to results of detections (text files, one per image)

• PREPROCESS_OUT_DIR - path to a directory containing preprocessed images.

• PREPROCESSED_FILES - path to a file containing list of images to be processed with their original sizes file.mame;width;height. This file contains only image file names without paths.

• IMAGE_COUNT - count of images which will be taken from dataset to process. Default value: 1.

• BATCH_SIZE - size of batch (number of images that can be processed "simultaneously"). Now batch processing by TF Lite is not supporting. Should be: 1

• NUMBER_OF_PROCESSORS - indicate how many threads TF Lite can use to process images. Default value: 1.

• FULL_REPORT - additional logging level (True or False)
• VERBOSE - maximal logging level (True or False)

Program-specific preprocessing

A client program may contain its own additional preprocessing script preprocess-next.py. If it exists, it will be called after the common preprocessing.

Images dataset

A client program should provide access to the dataset via run-time dependency dataset, e.g.:

  "run_deps": {
    "dataset": {
      "local": "yes",
      "name": "Object detection dataset",
      "sort": 20,
      "tags": "dataset,object-detection"
    },

Weights package

The model for benchmarking is provided by a client program's run-time dependency weights, e.g.:

  "run_deps": {
    "weights": {
      "force_target_as_host": "yes",
      "local": "yes",
      "name": "TensorFlow model",
      "sort": 60,
      "tags": "ssd,tflite,model,object-detection"
    }

Program parameters

Here we describe a client program's parameters affecting the pre/post-processig stages.

Input image parameters

CK_BATCH_COUNT

The number of processed images

Default: 1.

CK_SKIP_IMAGES

The number of images which will be skiped before processed images

Default: 0.

• 2.6.0 (2020-09-22)
• 2.5.0 (2020-09-04)
• 2.0.3 (2020-09-03)
• 1.1.1 (2020-09-02)
• 1.1.0 (2020-05-02)
• 1.0.0 (2019-10-17)

• .cm/desc.json (3 bytes)
• .cm/info.json (311 bytes)
• .cm/meta.json (3 bytes)
• README.md (4161 bytes)
• calc_metrics_coco_pycocotools.py (1820 bytes)
• calc_metrics_coco_tf.py (4284 bytes)
• calc_metrics_kitti.py (344 bytes)
• calc_metrics_oid.py (4964 bytes)
• ck_utils.py (3343 bytes)
• converter_annotations.py (4840 bytes)
• converter_results.py (4253 bytes)
• iniless_postprocess.py (6988 bytes)
• loadgen_postprocess.py (5200 bytes)
• loadgen_preprocess.py (6780 bytes)
• schindler.py (1752 bytes)