This folder uses purified and reordered dataset
This test also tried to see how g_loss changes through
- Dataset
Action class: kicking
xsub
integrated training and valid dataset
eliminated samples are
[3, 12, 13, 21, 23, 70, 76, 80, 86, 93, 115, 119, 123, 127, 136, 140, 155, 159, 161, 163, 165, 168, 172, 176, 179, 185, 193, 194, 199, 205, 207, 211, 213, 214, 217, 233, 234, 238, 257, 259, 263, 269, 276, 282, 285, 304, 321, 330, 331, 342, 346, 347, 352, 355, 361, 362, 363, 375, 394, 402, 427, 429, 437, 439, 441, 445, 448, 449, 466, 467, 468, 471, 479, 481, 484, 485, 522, 537, 539, 544, 553, 554, 555, 562, 566, 567, 568, 572, 578, 588, 589, 600, 601, 604, 608, 612, 613, 617, 630, 639, 641, 643, 644, 648, 649, 654, 658, 659, 665, 668, 669, 727, 732, 733, 737, 739, 741, 751, 762, 775, 784, 787, 789, 802, 804, 806, 815, 817, 819, 836, 838, 841, 854, 859, 868, 870, 882, 893, 898, 901, 913, 921, 925, 927]
Joints are rearranged in following order.
[20, 19, 18, 17, 24, 25, 12, 11, 10, 9, 4, 3, 21, 2, 1, 5, 6, 7, 8, 23, 22, 13, 14, 15, 16]
Test Result
-epoch 500까지-fluctuation은 전반적으로 적어진 듯 하다
-g_loss를 특정 간격을 두고 변화치를 관측
-하지만 loss가 솟아 오르는 부분이 이전 test보다 더 높게 솟는다. 이 부분의 sample을 우선 확인해볼 것.