No, when the one-shot command is used, that is the end of training. Showing multiple angles of animal does not enhance its accuracy in identification. Learning from multiple samples are the case for DNN.
Say, for example from the demo, a picture of a Parrot. The parrot was successfully learned using the existing background that was previously used to train tiger, elephant, car etc so by your logic, AKIDA had enough one-shots to subtract background from the object.
But it was still not able to recognize correctly as Parrot when shown a photo. It jumped all over the place from car, background, elephant etc.
As always, I stay within the factual observation, whether liked or not.
2020 BRN Discussion, page-18271
-
- There are more pages in this discussion • 11,043 more messages in this thread...
You’re viewing a single post only. To view the entire thread just sign in or Join Now (FREE)
Add BRN (ASX) to my watchlist
 (20min delay)
|
|||||
|
Last
17.5¢ |
Change
0.005(2.94%) |
Mkt cap ! $342.8M | |||
| Open | High | Low | Value | Volume |
| 17.5¢ | 18.0¢ | 17.0¢ | $1.233M | 7.023M |
Buyers (Bids)
| No. | Vol. | Price($) |
|---|---|---|
| 31 | 1279811 | 17.0¢ |
Sellers (Offers)
| Price($) | Vol. | No. |
|---|---|---|
| 18.0¢ | 1408579 | 46 |
View Market Depth
| Last trade - 16.10pm 14/08/2024 (20 minute delay) ? |
| BRN (ASX) Chart |