Name of the Log file, such as _100-587-10.1(7)-12.2(1)-d=1.log,
identifies
- preprosessing steps: 100 (using intensities only), 200 (using
intensities only, local-view illumination invariant), 111 (using also X anf Y
gradients)
- the network size 587 (= 24x24x(number of image modalities)+IDs)
- the scenario used for testing 10.1(7)-12.2(1)
10 videos. 1st of the pair (every 7th frame used) are used for memorization.
12 videos. 2nd of the pair (every frame
used) are used in recognition. NB: 2 face clips used for testing were not
memorized in this example.
- additional parameters: d=1
In memorization
C:\Cpp\video-DB\160x120\dmitry-short1.avi as 5 ... 7 M=98 5.64 E=1006 14 M=99 5.69 E=592 .... N=1739, M=108, <Cii>^2=0.0039, <Cij^2>=0.0000 |
Face in video is memorized with ID=5 ... Frame number 7, Total number of frames used (presented) so far to the network M=98 , its percentage wrt to the number of neurons (M/N) 5.64, and E=592, which is the distance from new vector to the The smaller it is, the more similar is the new stimulus to what has been already memorized. |
In recognition
C:\Cpp\video-DB\160x120\dmitry-short2.avi as 5 * 22 Ssum[1][5]=0.11 -1.0 -0.6 -1.2 -0.7 -0.7 +0.1 -0.5 -1.1 -1.1 -0.7 -1.5 ( 21 476 ) . 24 Ssum[1][1]=-0.58 -1.1 -0.6 -1.2 -0.8 -0.8 -0.3 -0.7 -1.3 -1.0 -0.5 -1.3 ( 16 468 ) * 26 Ssum[1][5]=0.15 -1.1 -1.0 -1.0 -0.6 -1.0 +0.2 -0.6 -1.2 -1.1 -0.8 -1.6 ( 19 473 )
* 70 Ssum[1][5]=0.41 -1.0 -0.5 -1.1 -0.3 -1.0 +0.4 -0.9 -1.2 -1.3 -1.1 -0.8 ( 16 493 ) + 72 Ssum[1][5]=0.07 -0.8 -0.1 -1.1 +0.2 -1.3 +0.1 -0.6 -0.9 -0.5 -0.9 -0.7 ( 19 505 ) . 74 Ssum[1][1]=-0.48 -1.1 -0.5 -1.0 -0.3 -1.3 -0.3 -1.0 -1.0 -1.0 -0.9 -0.8 ( 48 481 ) 76 Ssum[1][3]=0.37 -1.0 -0.7 -1.0 +0.4 -1.5 -0.6 -0.7 -0.7 -1.0 -0.8 -1.4 ( 31 502 ) + 82 Ssum[1][5]=0.07 -1.0 -1.2 -1.3 +0.3 -1.2 +0.1 -0.7 -0.9 -1.1 -1.0 -0.8 ( 16 501 )
*,+,.,(space) indicate the outcome based of the single frame: * - best case result (S10): + - S11 result: . - S00 result: - S01,S02 results:
Row of numbers show Postsynaptic potentials (PSP) of name tag neurons
(when the network reached attractor after having seen a visual stimulus of
one frame)
Based on these PSPs, the final decision on which name tag neuron wins
(i.e. what is answer to the question WHO IS IT?) is made.
There are several ways (techniques) to make this decision:
a) neural mode: all > 0 (all > S0) are considered as "win"s
b) max mode: maximal PSP wins
c) PSP time-filtered (averaged): technique of a) or b) is used on the averaged
over several consecutive PSPs instead of PSP of a individual frame
d) time-filtered (averaged): average or median of decisions which were made on
several consecutive individual frames (made according to a) or b))
and any combination of the above.
The last two numbers in brackets indicate number of iteration before reaching an attractor and E value.
The total of ALL S10, S11, S01, S00 ... S02 results are summarized in the end of the file as:
10 clips, 285 frames in training. 1210 in testing. Statistics: 10 11 01 00 | right, but all<S | wrong, but many>S ID 0 & 24 & 3 & 0 & 1 \\ \hline %4i 0 0 ID 1 & 87 & 0 & 1 & 7 \\ \hline %4i 0 0 ID 2 & 140 & 0 & 1 & 13 \\ \hline %4i 0 0 ID 3 & 79 & 1 & 4 & 49 \\ \hline %4i 0 0 ID 4 & 41 & 2 & 2 & 17 \\ \hline %4i 0 0 ID 5 & 105 & 0 & 1 & 6 \\ \hline %4i 0 0 ID 6 & 105 & 2 & 6 & 7 \\ \hline %4i 0 0 ID 7 & 180 & 1 & 5 & 15 \\ \hline %4i 0 0 ID 8 & 90 & 4 & 13 & 38 \\ \hline %4i 0 7 ID 9 & 126 & 0 & 5 & 13 \\ \hline %4i 0 9 Total:& 977 & 13 & 38 & 166 \\ \hline %4i 0 16 Network: N=1739 (00001-111), nIDs=11 || D=0.15, T=1, S0=0.00
For more details: see publications
Created: 10.XII.2004.
Last Updated: 14.IV.2005
Computational Video Group, IIT-ITI,
NRC-CNRC
Project Leader:
Dmitry
O. Gorodnichy.
Email for sending comments: memory@perceptual-vision.com
www.perceptual-vision.com
(synapse.vit.iit.nrc.ca).
Copyright
© 2004-2005