Skip to content

[Feature]: Represent in SDS Metadata file a single-sample block with variable size. #300

Description

@ReinhardKeil

Arm-Examples/ModelNova#32 introduces a different data format.

Data that we write to the SDS file (and process with SDS-View and SDS VS Code extension) are:
• 4 bytes (float): probability that detected object was PAPER
• 4 bytes (float): probability that detected object was ROCK
• 4 bytes (float): probability that detected object was SCISSORS
• 4 bytes (float): probability that detected object was UNKNOWN

Data they write if flag F was activated and that they need for processing with Fusion Studio are:
• 100 bytes (null terminated string): label name of predicted class name
• 2 bytes (uint16_t): predicted class index
• 2 bytes: padding
• 4 bytes (float): confidence score of the predicted class

In case of an active F flag, the SDS utilities cannot interpret the data.

Potential solutions imply that a dynamic block size is interpreted as a single-sample block.

Extend Metadata format with:

  • (A) option to represent dynamic size information (i.e. dim-x: *). This could be only specified in the last value of a content: node.
  • (B) indicate that a block contains only a single sample. The rest of the content of a block would be then ignored.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions