Pouring, Naive Baselines (Pointwise with Classification PN++)

Why is this baseline so awful? We really have to investigate, I think.

Created on June 4|Last edited on June 6

Comment

﻿
See results from the baseline of classification PN++ with pointwise loss. These look awful due to exploding MSEs, but it was overfitting to 1 demonstration? I'm convinced there's some kind of bug, despite spending a day doing what I thought was fixing the bugs. :(
Compare with naive methods:
﻿https://wandb.ai/mooey5775/mixed_media/reports/Pouring-Naive-Vector-Methods-data-v02--VmlldzoyMTAxOTUw 
Section 1﻿
eval/info_done_final
eval/info_done_final
050100150200250Step00.20.40.60.81
BC03_PourWater_v01_ntrain_0100_PCL_PNet2_classif_6D_pointwise_ee2flow_3DoF_ar_8_hor_100_scalePCL_noScaleTarg_2022_06_03_10_14_28_0003
BC03_PourWater_v01_ntrain_0100_PCL_PNet2_classif_6D_pointwise_ee2flow_3DoF_ar_8_hor_100_scalePCL_noScaleTarg_2022_06_03_10_14_28_0002
BC03_PourWater_v01_ntrain_0100_PCL_PNet2_classif_6D_pointwise_ee2flow_3DoF_ar_8_hor_100_scalePCL_noScaleTarg_2022_06_03_10_36_23_0002
BC03_PourWater_v01_ntrain_0100_PCL_PNet2_classif_6D_pointwise_ee2flow_3DoF_ar_8_hor_100_scalePCL_noScaleTarg_2022_06_03_10_36_23_0001
BC03_PourWater_v01_ntrain_0100_PCL_PNet2_classif_6D_pointwise_ee2flow_3DoF_ar_8_hor_100_scalePCL_noScaleTarg_2022_06_03_10_14_28_0001
eval/info_done_final
eval/info_done_final
050100150200250Step00.20.40.60.81
Group
Classification PN++ to Vector, Pointwise5
 
Classification PN++ to Vector, MSE5
﻿
﻿

Add a comment