What is a Blob and How to Extract Detection Data

Unable to determine how to use a “blob” when working with the OpenVINO™ toolkit and how to extract detection output data.

In older OpenVINO™ releases (Inference Engine–based APIs), a “Blob” referred to a container for input or output data passed to the inference engine. This concept was used with APIs such as SetBlob().

In current OpenVINO™ releases, the term “Blob” is no longer used. Instead, OpenVINO works directly with tensors:

• In C++, data is represented using ov::Tensor |.
• In Python*, data is typically handled as NumPy arrays.

Detection data is extracted by accessing the output tensors of a compiled model and interpreting them according to the model’s output layer definition. Output size and structure are determined by the model topology and output layers, not by the size of the .bin file.

For users migrating legacy code:

• Replace Blob and SetBlob() usage with tensor-based input/output handling.
• Use named outputs or output indices from the compiled model to retrieve detection results.

The “Blob” terminology originates from legacy Inference Engine APIs and may still appear in older samples or documentation. For current OpenVINO workflows, refer to the OpenVINO Runtime documentation, which describes tensor-based input/output handling and detection result parsing.

Refer to OpenVINO™ Integrations for a high-level description of the process of integrating OpenVINO™ into your application.