Article ID: 000089260 Content Type: Error Messages Last Reviewed: 05/20/2022

Error: “Unknown Dataset Type” When Importing Custom COCO Dataset into OpenVINO™ Deep Learning Workbench (DL Workbench)

BUILT IN - ARTICLE INTRO SECOND COMPONENT
Summary

Correct COCO dataset structure to use in OpenVINO™ DL Workbench.

Description
  • Converted a custom model to Intermediate Representation (IR) format with precision FP32.
  • Converted IR (FP32) to IR (INT8) with a not-annotated dataset and default configuration.
  • Bad result when inferencing with IR (INT8).
  • Created JSON file with annotations with the following format:

    {
    "images": [<images>],
    "annotations": [<annotations>]
    }

  • Created custom COCO dataset.
  • Failed to import custom COCO dataset to OpenVINO™ DL Workbench:

Error occurred while uploading dataset: Unknown dataset type

Resolution

OpenVINO™ DL Workbench recognizes the annotated dataset as COCO dataset format by checking the JSON file.

  • Create .JSON file containing the following entries:

    {
    "info": <info>,
    "images": [<images>],
    "licenses": [<licenses>],
    "annotations": [<annotations>]
    }

  • Add "info" and "licenses" entries as empty list if not already available in the JSON file.
  • Create COCO dataset with the following format:

    |-- val
    |-- 0001.jpg
    |-- 0002.jpg
    ...
    |-- n.jpg
    |-- annotations
    |-- instances_val.json

Additional information

Refer to Dataset Types for more information regarding the various dataset use cases and dataset structures for DL Workbench.

Related Products

This article applies to 2 products