BC03 MMOneSphere Ablation No Skip Connections

Created on June 14|Last edited on June 16
Comment
Now with 3D flow.
This only has 3X random seeds, I need to find time / compute to do more later (but that will be for BC04). Actually the results look good but it's always high variance and with 5 seeds it should lower considerably.
Update: ugh, the cluster crashed, so the 2 runs here are not finished. But objectively if you look at the smoothed results from the 3 that ran to completion it's 0.4091 which is lower than 0.4711 we got with the proposed method (well that was 6D flow). Actually it's barely worse than the 0.4095 we got with 3D flow! So I honestly think the method is quite comparable, which is interesting to see with no skip connections!
Results﻿
eval/info_done_final
eval/info_done_final
050100150200250Step00.20.40.60.81
BC03_MMOneSphere_v01_ntrain_0100_PCL_PNet2_svd_pointwise_ee2flow_4DoF_ar_8_hor_100_NOSKIP_scalePCL_noScaleTarg_2022_06_14_14_44_46_0002
BC03_MMOneSphere_v01_ntrain_0100_PCL_PNet2_svd_pointwise_ee2flow_4DoF_ar_8_hor_100_NOSKIP_scalePCL_noScaleTarg_2022_06_14_14_44_46_0001
BC03_MMOneSphere_v01_ntrain_0100_PCL_PNet2_svd_pointwise_ee2flow_4DoF_ar_8_hor_100_NOSKIP_scalePCL_noScaleTarg_2022_06_12_18_32_19_0003
BC03_MMOneSphere_v01_ntrain_0100_PCL_PNet2_svd_pointwise_ee2flow_4DoF_ar_8_hor_100_NOSKIP_scalePCL_noScaleTarg_2022_06_12_18_32_19_0002
BC03_MMOneSphere_v01_ntrain_0100_PCL_PNet2_svd_pointwise_ee2flow_4DoF_ar_8_hor_100_NOSKIP_scalePCL_noScaleTarg_2022_06_12_18_32_19_0001
eval/info_done_final
eval/info_done_final
050100150200250Step00.20.40.60.81
Group
Remove skip connections5
﻿
﻿
Add a comment