btgraham
commented
3 years ago It should be training accuracy, but I get (first 32 epochs)
test epoch 1 0.4044794946674115 MegaMulAdd= 1236.0365977761007 MegaHidden 9.882867823899371 time= 137.9610505104065 s
test epoch 2 0.4531207734074015 MegaMulAdd= 1235.7378010163523 MegaHidden 9.882053534591194 time= 135.87021899223328 s
test epoch 4 0.4698866501567467 MegaMulAdd= 1235.4892405534592 MegaHidden 9.88099307672956 time= 141.1514275074005 s
test epoch 8 0.5121569170531977 MegaMulAdd= 1235.4140991597483 MegaHidden 9.882710842767297 time= 134.76471185684204 s
test epoch 16 0.5468680402580118 MegaMulAdd= 1235.2842045987422 MegaHidden 9.881746495597483 time= 134.37742066383362 s
test epoch 32 0.557365473337408 MegaMulAdd= 1235.985770183648 MegaHidden 9.883132638993711 time= 134.7175693511963 s
Hello. Thank you so much for this work. I had a question.
https://github.com/facebookresearch/SparseConvNet/blob/master/examples/nyu2/train.py/#L164 Does stats['c']/stats['n'] mean accuracy? If it does, I am getting around 23% accuracy after 200 epochs.
Thank you again.