GrumpyZhou
commented
2 years ago Hi @handsomelcj ,
This was in my previous logging file and hope this solves your question:
------------ Configuration -------------
backbone:ResNet34
batch:4
change_stride:True
ckpt:None
cls_dthres:[50, 5]
conv_dims:[512, 512]
conv_kers:[3, 3]
conv_strs:[2, 1]
cthres:0.5
data_root:data
dataset:MegaDepth
epi_dthres:[50, 5]
epochs:25
fc_dims:[512, 256]
feat_comb:pre
feat_idx:[0, 1, 2, 3]
freeze_feat:87
gpu:0
ksize:2
lr_decay:['multistep', '0.2', '5']
lr_init:0.0005
match_npy:megadepth_pairs.ov0.35_imrat1.5.pair500.excl_test.npy
out_dir:output/patch2pix/sval.new_env.Mega.ov0.35.pair500.excl_test.freeze87.cs.pretrain/ks2fe0123ep50-5cls50-5_wcls10.0wepi1-1.lr0.0005lrms0.2-5/pre400_conv33dim512-512str2-1fc512-256_psz16-16a8
pair_root:data_pairs
panc:8
plot_counts:20
prefix:sval.new_env
pretrain:pretrained/ncn_ivd_5ep.pth
pshift:8
psize:[16, 16]
ptmax:400
regr_batch:1200
resume:True
save_step:1
seed:1
shared:False
visdom_host:atcremers71
visdom_port:9333
weight_cls:10.0
weight_decay:0
weight_epi:[1, 1]
----------------------------------------
...
>Epoch:9 Skipped:1 Loss:pair=15.61 cls_mid=0.53 cls_fine=0.40 epi_mid=5.67 epi_fine=0.61
Cls_mid:rec=0.68 prec=0.73 spec=0.75 acc=0.71 f1=0.69
Cls_fine:rec=0.79 prec=0.81 spec=0.87 acc=0.83 f1=0.79
Match:cmid_gt=17.88 mmid_gt=5.67 mfid_gt=1.10 ffid_gt=0.61
Epoch training time: 71442.49s
Thank you for your outstanding work. My question is how many of the four kinds of loss (cls_mid/cls_fine/epi_mid/epi_fine) finally converge to? And how long the training will take？