So far we have put together a data set of approximately 1100 bear face chips, all from Brooks Falls bears. The face chips were extracted from photos from the following sources:

Below is a table of the Brooks Falls bear chip data. We have face chips for 43 different bears. The table has a row for each bear. The columns are:

  • ID: the bear’s identification number
  • Nickname: the bear’s nickname (if it has one)
  • Count: the number of face chips we have for the bear
  • Example Face Images: a few example face chips for the bear
ID Nickname Count Example Face Images
032 Chunk 16
051 Diver Jr. 1
068   7
083 Wayne Brother 21
089 Backpack 28
094   11
128 Grazer 61
132   29
151 Walker 67
171   3
201   1
218 Ugly 3
261   13
273   43
274 Overflow 4
284   16
289   51
402   8
409 Beadnose 73
410 Four-Ton 77
415   1
435 Holly 4
469   3
480 Otis 131
482 Brett 23
489 Ted 2
500 Indy 5
503 Cubadult 54
505   21
634 Popeye 16
700 Marge 4
708 Amelia 2
719 Princess 19
744 Dent 2
747   20
755 Scare D Bear 13
775 Lefty 178
813 Nostril Bear 3
814 Lurch 11
818   1
854 Divot 3
856   36
868 Wayne Brother 7

What We Need

As you can see, there are quite a few bears for which we only have a couple images. We really need more images of these bears to use them for training. We think we need at least 20 good face chips per bear, preferably from different seasons and different times of the season. We’re hoping the set we get from Katmai will fill out our set, but we can always you more!

We are also not sure if we can use the images where the bear is looking far to the left or right or down. We’re planning a few experiments to see how the various poses effect our results. If we have to filter out some of these more extreme poses, our dataset will get even smaller.

If you have decent quality photos or videos of the Brooks Falls bears, or of any other brown bears, please contact us at bearid [at] hypraptive [dot] com.

We are not using any captures from the explore.org bearcams at this time, as we don’t think the quality is sufficient. Hopefully, once we have the application working reasonably well, we can adapt it to work with the bearcams as well!