Vision ANN/SNN Trainer - browser edition, v004 (real video)

Trains both models on this device's camera. Companion to firmware-v004.ino, which reads its class list and settings straight from config.json - no more copying #defines by hand.

Camera

Camera: not enabled

Resolution & Storage

Any square size works - try 32, 64, 96, 128, 160... higher resolutions train slower (roughly with the square of the size) but may resolve finer detail; lower resolutions are much faster to iterate with, especially for the SNN.

Storage: in-browser (IndexedDB) - use Export to get real files.

Classes (tap a name to collect)

Train & Infer

ANN

SNN

Model file transfer (to/from the device's SD card)

Filenames match the device exactly: myWeights.bin and mySnnWeights.bin, in the SD card's /header/ folder. A weight file only loads if its size matches the CURRENT resolution/channels/ class count exactly - same safety check the device uses.

Freeze layers & tunables
Device settings (read automatically by firmware-v004.ino from config.json)

Nothing to paste - the firmware reads this straight from /header/config.json on the SD card. This is just a summary to confirm the device will see what you expect.

Console