pcr-upm / opal23_headpose

Headpose estimation using OPAL (2023)
MIT License
43 stars 4 forks source link

Training and testing a model using Panoptic data set #3

Closed bobetocalo closed 1 year ago

bobetocalo commented 1 year ago

Training: panoptic_ann_train.txt Testing: panoptic_ann_test.txt

42439 frames

-m test -p alignment -a /home/database/alignment/faces/headpose/panoptic-toolbox/panoptic_ann_test.txt --alignment opal23_headpose --database panoptic --gpu 0 --rotation-mode euler --target-dist 1.6

MAE: 11.97036283150434
[yaw, pitch roll]: [ 8.2814807  14.66706747 12.96254032]
GE: 12.222820714766549
 > [0,60]:
MAE: 5.585357649797939
[yaw, pitch roll]: [6.40890512 6.48402922 3.86313861]
GE: 10.884288471197534
 > [60,120]:
MAE: 18.431902039120885
[yaw, pitch roll]: [ 7.64249487 24.30409868 23.34911256]
GE: 12.049072702640927
 > [120,180]:
MAE: 11.939657277654694
[yaw, pitch roll]: [10.77443581 13.28895609 11.75557993]
GE: 13.725380094913648
Confusion matrix:
[[ 341  298   14   12    2    2    6    8    4    0    0    0    0]
 [ 276 2681  847   87   26   24   20    9    5    0    1    0    0]
 [  24  567 2374  674   63   10   12    6    1    2    0    0    0]
 [  13   34  440 2474  475   19    8    5    3    1    0    1    0]
 [   5   17   40  685 2335  531   41   10    4    2    1    1    0]
 [   3   15   14   48  538 1828  581   20    6    4    3    3    2]
 [   6    8   13    8   41  764 2186  336   19    9    7    2    2]
 [   3    1    5    7    5   26  611 2229  490   34   14   11    3]
 [   0    1    7    2    6   11   47  696 2669  733   46   17    6]
 [   4    2    2    2    5   14   19   41  953 2187  533   22    5]
 [   2    2    2   10    7    1    8    6   50  555 2144  727   17]
 [   0    0    1    0    2   10   14   17   15   55  667 2952  400]
 [   0    0    0    0    0    0    2    3    2    8   32  872  380]]
Accuracy (accurate class): 63.102%
Accuracy (adjacent class): 96.678%

-m test -p alignment -a /home/database/alignment/faces/headpose/panoptic-toolbox/panoptic_ann_test.txt --alignment opal23_headpose --database panoptic --gpu 0 --rotation-mode quaternion --target-dist 1.6

MAE: 10.46064735974909
[yaw, pitch roll]: [ 6.71373345 12.92472621 11.74348242]
GE: 10.197335613831632
 > [0,60]:
MAE: 4.363986957158695
[yaw, pitch roll]: [4.54789014 5.17503676 3.36903397]
GE: 7.959065532144632
 > [60,120]:
MAE: 14.914925088794945
[yaw, pitch roll]: [ 5.50362526 19.98000131 19.2611487 ]
GE: 8.650826046489309
 > [120,180]:
MAE: 12.125094050649196
[yaw, pitch roll]: [10.06220748 13.66480728 12.64826739]
GE: 13.949976337219375
Confusion matrix:
[[ 408  249   20    4    3    1    0    2    0    0    0    0    0]
 [ 225 3066  569   58   21   15   15    2    3    1    1    0    0]
 [   4  572 2554  556   36    3    6    1    1    0    0    0    0]
 [   4   20  344 2628  458   10    2    3    2    2    0    0    0]
 [   1   16   59  591 2459  507   20    7    7    5    0    0    0]
 [   1   14   23   64  513 1938  451   14   17   13    7    9    1]
 [   2   13   19   36   71  599 2139  425   38   29   20    9    1]
 [   0    3    6   10   15   16  293 2532  476   51   18   16    3]
 [   0    1    2    0    3    1    9  596 2986  540   61   39    3]
 [   1    0    0    0    2    2   12   15  737 2469  497   46    8]
 [   0    1    1    2    1    1    1    3   20  422 2471  598   10]
 [   0    1    1    1    1    0    1    5   18   47  555 3158  345]
 [   0    1    0    0    0    0    0    2    2    3   19  566  706]]
Accuracy (accurate class): 69.545%
Accuracy (adjacent class): 97.076%

-m test -p alignment -a /home/database/alignment/faces/headpose/panoptic-toolbox/panoptic_ann_test.txt --alignment opal23_headpose --database panoptic --gpu 0 --rotation-mode 6d --target-dist 1.6

MAE: 9.148617959716665
[yaw, pitch roll]: [ 5.74869927 11.48752183 10.20963278]
GE: 8.74636142722818
 > [0,60]:
MAE: 4.319680508511525
[yaw, pitch roll]: [4.51272383 5.24765083 3.19866687]
GE: 7.856118640611266
 > [60,120]:
MAE: 14.57160335857428
[yaw, pitch roll]: [ 5.26494227 19.50103711 18.94883069]
GE: 8.231639254145728
 > [120,180]:
MAE: 8.59601049374029
[yaw, pitch roll]: [7.45536671 9.7801491  8.55251567]
GE: 10.139813043705155
Confusion matrix:
[[ 442  219   13    3    4    3    1    2    0    0    0    0    0]
 [ 254 3146  492   34   13   12   10    8    4    3    0    0    0]
 [   4  580 2524  586   23    5    3    7    1    0    0    0    0]
 [   1   10  316 2664  452   20    3    3    4    0    0    0    0]
 [   0    1   24  600 2559  440   27    8    6    4    1    2    0]
 [   1    3   14   21  473 1870  650   20    5    5    3    0    0]
 [   0    0    4    7   16  420 2479  451   11    8    3    2    0]
 [   0    0    1    2   11   12  268 2604  509   21    7    3    1]
 [   0    0    0    0    0    1    7  519 3138  545   21   10    0]
 [   0    0    1    1    5    1    3   25  737 2492  501   17    6]
 [   0    0    1    2    1    1    1    7   22  376 2573  539    8]
 [   0    0    0    0    1    1    0    4    6   19  534 3232  336]
 [   0    0    0    0    0    0    1    0    3    3   23  549  720]]
Accuracy (accurate class): 71.734%
Accuracy (adjacent class): 98.468%

-m test -p alignment -a /home/database/alignment/faces/headpose/panoptic-toolbox/panoptic_ann_test.txt --alignment opal23_headpose --database panoptic --gpu 0 --rotation-mode 6d_opal --target-dist 1.6

MAE: 8.798797672930316
[yaw, pitch roll]: [ 5.55258397 11.11441056  9.7293985 ]
GE: 8.680376661697201
 > [0,60]:
MAE: 4.065320877771171
[yaw, pitch roll]: [4.1214775  5.04090342 3.03358171]
GE: 7.431575717351682
 > [60,120]:
MAE: 13.438788015190616
[yaw, pitch roll]: [ 5.04808815 17.95497797 17.31329792]
GE: 8.095129468157763
 > [120,180]:
MAE: 8.92435888996843
[yaw, pitch roll]: [ 7.47376218 10.39972304  8.89959145]
GE: 10.500004187597003
Confusion matrix:
[[ 437  221   12    6    5    1    2    2    1    0    0    0    0]
 [ 182 3151  548   49   15   10    8    5    2    4    1    1    0]
 [   5  516 2630  559   13    0    5    1    2    1    0    1    0]
 [   1   15  335 2682  424    7    4    3    1    1    0    0    0]
 [   0    8   37  491 2649  450   21    5    7    1    2    0    1]
 [   0    4    9   25  444 1938  614   13   11    5    2    0    0]
 [   0    3    5   10   25  396 2600  342   12    6    0    2    0]
 [   0    1    2    6    5   10  377 2586  411   26   10    5    0]
 [   0    0    0    0    0    2   10  607 3004  539   52   25    2]
 [   0    1    2    3    2    3    4    7  691 2615  431   28    2]
 [   0    1    1    0    0    2    2    8   10  392 2636  471    8]
 [   0    0    0    0    1    0    8   15   12   32  559 3207  299]
 [   0    0    0    0    0    0    1    2    7    2   11  556  720]]
Accuracy (accurate class): 72.704%
Accuracy (adjacent class): 98.282%