search

bcmi / GracoNet-Object-Placement

Official code for ECCV2022 paper: Learning Object Placement via Dual-path Graph Completion
MIT License
73 stars 6 forks source link

训练时这样的情况是否是收敛的 #4

Closed lzclzclzc12 closed 1 year ago

lzclzclzc12 commented 1 year ago

[Epoch 11/11] [Batch 3811/3880] [G - gan1: 6.830, gan2: 6.101, rec: 0.151, kld: 0.0976] [D - real: 0.428, fake1: 0.016, fake2: 0.023] [Epoch 11/11] [Batch 3821/3880] [G - gan1: 6.830, gan2: 6.102, rec: 0.151, kld: 0.0978] [D - real: 0.428, fake1: 0.016, fake2: 0.023] [Epoch 11/11] [Batch 3831/3880] [G - gan1: 6.833, gan2: 6.104, rec: 0.151, kld: 0.0980] [D - real: 0.428, fake1: 0.016, fake2: 0.023] [Epoch 11/11] [Batch 3841/3880] [G - gan1: 6.835, gan2: 6.107, rec: 0.151, kld: 0.0982] [D - real: 0.428, fake1: 0.016, fake2: 0.023] [Epoch 11/11] [Batch 3851/3880] [G - gan1: 6.838, gan2: 6.110, rec: 0.151, kld: 0.0984] [D - real: 0.428, fake1: 0.016, fake2: 0.023] [Epoch 11/11] [Batch 3861/3880] [G - gan1: 6.839, gan2: 6.114, rec: 0.151, kld: 0.0986] [D - real: 0.428, fake1: 0.016, fake2: 0.022][Epoch 11/11] [Batch 3871/3880] [G - gan1: 6.840, gan2: 6.116, rec: 0.151, kld: 0.0988] [D - real: 0.428, fake1: 0.016, fake2: 0.022]

Siyuan-Zhou commented 1 year ago

@lzclzclzc12 The training losses might increase in some periods and decrease in other periods, which can not precisely reflect the training quality. We suggest using evaluation results for a more straightforward reflection.