Export Project

You can download an archive with artifacts of your project:

A downloaded archive also contains information on the project and best model performance.

The archive is organized as follows:

|-- configs
  |-- accuracy_config.yml
  |-- calibration_config.json
|-- dataset
  |-- dataset.tar.gz
|-- model
  |-- model.xml
  |-- model.bin
|-- description.txt

The example for the mobilenet-ssd model with a COCO dataset.

- name: mobilenet-ssd
  - framework: dlsdk
    device: CPU
      type: ssd
    batch: 1
    model: /home/workbench/.workbench/models/3/16/mobilenet-ssd.xml
    weights: /home/workbench/.workbench/models/3/16/mobilenet-ssd.bin
  - name: dataset
    data_source: /home/workbench/.workbench/datasets/2
      annotation_file: /home/workbench/.workbench/datasets/2/instances_val2017_200pictures.json
      converter: mscoco_detection
      images_dir: /home/workbench/.workbench/datasets/2
    subsample_size: 100%
    - type: resize
      size: 300
    - type: resize_prediction_boxes
    - type: map
      integral: 11point
      ignore_difficult: true
      presenter: print_vector

The example is for the mobilenet-ssd model calibrated on a COCO dataset with Default method.

   "model": {
      "model_name": "mobilenet-ssd",
      "model": "/home/workbench/.workbench/models/3/original/mobilenet-ssd.xml",
      "weights": "/home/workbench/.workbench/models/3/original/mobilenet-ssd.bin"
   "engine": {
      "name": "mobilenet-ssd",
      "launchers": [
            "framework": "dlsdk",
            "device": "CPU",
            "adapter": {
               "type": "ssd"
            "batch": 1
      "datasets": [
            "name": "dataset",
            "data_source": "/home/workbench/.workbench/datasets/2",
            "annotation_conversion": {
               "annotation_file": "/home/workbench/.workbench/datasets/2/instances_val2017_200pictures.json",
               "converter": "mscoco_detection",
               "images_dir": "/home/workbench/.workbench/datasets/2"
            "subsample_size": "100%",
            "preprocessing": [
                  "type": "resize",
                  "size": 300
            "postprocessing": [
                  "type": "resize_prediction_boxes"
            "metrics": [
                  "type": "map",
                  "integral": "11point",
                  "ignore_difficult": true,
                  "presenter": "print_vector"
   "compression": {
      "target_device": "CPU",
      "algorithms": [
            "name": "DefaultQuantization",
            "params": {
               "preset": "performance",
               "stat_subset_size": 60

The example of description.txt with information on the project, best model performance, corresponding performance reports, and the execution graph of the model on the device that you used in the project.

Model: mobilenet-ssd
Dataset: coco200
Device: CPU | Intel(R) Core(TM) i5-7300U CPU @ 2.60GHz
Target: local | | Local Workstation
Optimized with INT8 Calibration: Yes
Optimized with Winograd Algorithmic Tuning: No
Corresponding latency: <latency value>
Best result FPS: <FPS value>
Best result batch configuration: 1
Best result stream configuration: 1
DL Workbench version: 2021.4.2

Accuracy Checker version: 0.8.6
Post-training Optimization Tool version: 1.0

If you submit a ticket about model support to the OpenVINO™ team, it is highly recommended to attach the model and its execution graph, which you can obtain on this tab.

To export a project, follow the steps below.

  1. Go to the Perform tab on the Projects page. Open the Export Project subtab.

  2. Check the items you want to include. You cannot export project-related information without including at least one element to the package.

  3. Click Export. The exporting process starts on the server followed by an automatic archive download.

Below is an example of a project being ready for export. In this example, the model, accuracy configuration, and calibration configuration will be included in the export package, while the package will not contain the dataset.