Skip to content

fix(app): tidy class labels and input shapes on application cards - #62

Merged
nikunj95 merged 4 commits into
mainfrom
fix/bcc-hd-label
Oct 30, 2026
Merged

nikunj95 merged 4 commits into
mainfrom
fix/bcc-hd-label

Conversation

@nikunj95

@nikunj95 nikunj95 commented Oct 30, 2026 •

Copy link
Copy Markdown
Collaborator

Three changes to the More Information section of an application card, both from the captain's testing of the keyword spotting and human detection demos.

A heading that fits the demo. The Vision Human Detection card listed its class label under a Keywords heading ("Keywords: person"). That heading fits keyword spotting, where the labels are the words the board listens for, but the vision demo only says whether a person is in view. The board's details reply carries each model's class labels (already without silence and unknown); the app put every application's labels under one hard-coded heading. Each application can now name that field itself:

  • Vision Human Detection: Label: with Person detected. The board sends person in lowercase, so the app capitalizes it.
  • Keyword Spotting and any other application: Keywords: with the labels as the board sends them.

Labels on the heading's line. The labels used to sit on a row of their own under the heading. They now start on the heading's line, so the vision card reads Label: Person detected on one line. The heading and its labels share one wrapping line, so when keyword spotting's ten labels wrap, the following rows start at the left edge of the word Keywords and use the card's full width. This looks the same on the BrainBoard1500 and the AkidaTag, which report the same ten keywords.

One way to write an input shape. The AkidaTag reports its keyword model's input shape as 49 x 10 x 1, while the BrainBoard1500 reports 49x10x1 and 96x96x3. The card now drops the spaces, so every board's shape reads like 49x10x1. __tests__/formatInputShape.test.ts covers both spellings.

This changes only the app; the board firmware and its wire format are untouched. __tests__/applicationClassLabels.test.tsx opens More Information on both cards and pins each heading together with its labels on one line.

How to test

  1. Install a BrainChip Connect build that includes this change.
  2. Turn on the Nicla Vision with the BrainBoard 1500 running the NiclaVisionConnect sketch.
  3. In the app, tap BrainBoard1500, then Connect to Device.
  4. On Select the Application, tap More Information on the Vision Human Detection card.
    • Pass: Input Shape: reads 96x96x3, and the last line reads Label: with a Person detected chip on the same line. The word "Keywords" does not appear on this card.
  5. Tap More Information on the Keyword Spotting card.
    • Pass: Input Shape: reads 49x10x1. Keywords: is followed on the same line by the first chips (down, go, left). The rows that wrap below start at the left edge of the word Keywords and use the card's full width (no, off, on, right, stop, then up, yes). All ten keywords are there, with no silence or unknown.
  6. Go to Profile, tap Unpair Device, then Unpair, Disconnect and OK.
  7. Turn on the AkidaTag, tap AkidaTag, then Connect to Device. Only the Keyword Spotting card is listed.
  8. Tap More Information on it.
    • Pass: Input Shape: reads 49x10x1, with no spaces, although the AkidaTag reports it as 49 x 10 x 1. The keywords have the same layout as step 5: the keywords start on the Keywords: line and the wrapped rows start at the left edge of the word Keywords.

The application card listed the Human Detection model's class labels
under a Keywords heading, which only fits keyword spotting. Each
application can now name its class label field: the vision card shows
Label: Person detected, and keyword spotting keeps its Keywords list.
More Information put the class labels on a row of their own under the
Keywords or Label heading. They now start on the heading's line, and
when keyword spotting's labels wrap, the following rows start under the
first label rather than back under the heading.
@nikunj95 nikunj95 changed the title fix(app): stop calling what the vision demo detects keywords fix(app): put class labels beside a heading that fits the demo Oct 30, 2026
The captain chose wrapped keyword rows that start where the Keywords
heading starts, so each row after the first uses the card's full width.
The labels now sit in the heading's own wrapping line instead of a
nested row beside it.
The AkidaTag reports a keyword model's input shape as 49 x 10 x 1 while
the BrainBoard1500 reports 49x10x1. More Information now drops the
spaces, so every board's shape reads like 49x10x1 or 96x96x3.
@nikunj95 nikunj95 changed the title fix(app): put class labels beside a heading that fits the demo fix(app): tidy class labels and input shapes on application cards Oct 30, 2026
@nikunj95
nikunj95 merged commit f21d18d into main Oct 30, 2026
5 checks passed
@nikunj95
nikunj95 deleted the fix/bcc-hd-label branch October 30, 2026 23:13
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant