finetune nan

[lijain]

感谢你开源的代码。我在finetune时在第9轮时图像还是崩的，下面我放上我的log和生成结果图，请问下是什么原因
这个是我加载权重的后的log信息，
0e_10iter: perceptual - 232.17307; gen_gan - 1.34401; feature_matching - 15.46109; equivariance_value - 0.29036; keypoint - 15.23777; headpose - 112.23193; expression - 1.34486; disc_gan - 0.21649
0e_20iter: perceptual - 219.37776; gen_gan - 1.48003; feature_matching - 15.35167; equivariance_value - 0.29880; keypoint - 15.21261; headpose - 94.12276; expression - 1.35599; disc_gan - 0.11738
0e_30iter: perceptual - 212.06287; gen_gan - 1.46152; feature_matching - 15.39599; equivariance_value - 0.31251; keypoint - 15.18884; headpose - 96.09579; expression - 1.30995; disc_gan - 0.19136
0e_40iter: perceptual - 210.74174; gen_gan - 1.56604; feature_matching - 15.41683; equivariance_value - 0.28903; keypoint - 15.19454; headpose - 98.82621; expression - 1.32057; disc_gan - 0.11889
0e_50iter: perceptual - 207.96489; gen_gan - 1.54837; feature_matching - 15.37361; equivariance_value - 0.28908; keypoint - 15.17129; headpose - 95.09063; expression - 1.29430; disc_gan - 0.13431
0e_60iter: perceptual - 204.52927; gen_gan - 1.65631; feature_matching - 15.44373; equivariance_value - 0.30531; keypoint - 15.19445; headpose - 91.40559; expression - 1.28033; disc_gan - 0.11218
0e_70iter: perceptual - 203.76723; gen_gan - 1.62658; feature_matching - 15.39420; equivariance_value - 0.29203; keypoint - 15.19353; headpose - 98.69791; expression - 1.23485; disc_gan - 0.08718
0e_80iter: perceptual - 203.75139; gen_gan - 1.68678; feature_matching - 15.46345; equivariance_value - 0.28991; keypoint - 15.19819; headpose - 93.80473; expression - 1.36832; disc_gan - 0.11625
0e_90iter: perceptual - 198.74699; gen_gan - 1.70415; feature_matching - 15.47408; equivariance_value - 0.30915; keypoint - 15.18057; headpose - 88.48777; expression - 1.28243; disc_gan - 0.10653