github/DeepFaceLive
1
0
Fork 0
mirror of https://github.com/iperov/DeepFaceLive synced 2025-08-19 21:13:21 -07:00
Code Issues Projects Releases Packages Wiki Activity Actions

_

Browse source
This commit is contained in:
iperov 2021-11-13 18:15:54 +04:00
commit 97991ca5cf
2 changed files with 8 additions and 13 deletions
Download patch file Download diff file Expand all files Collapse all files

8
apps/trainers/FaceAligner/FaceAlignerTrainerApp.py
View file

@ -369,9 +369,11 @@ class FaceAlignerTrainerApp:
d = len(lh) // max_lines d = len(lh) // max_lines
lh_ar = np.array(lh[-d*max_lines:], np.float32) lh_ar = np.array(lh[-d*max_lines:], np.float32)
lh_ar = lh_ar.reshape( (max_lines, d)).mean(-1) lh_ar = lh_ar.reshape( (max_lines, d))
lh_ar_max, lh_ar_min, lh_ar_mean, lh_ar_median = lh_ar.max(-1), lh_ar.min(-1), lh_ar.mean(-1), np.median(lh_ar, -1)
print( '\n'.join( f'{value:.5f}' for value in lh_ar ) )
print( '\n'.join( f'max:[{max_value:.5f}] min:[{min_value:.5f}] mean:[{mean_value:.5f}] median:[{median_value:.5f}]' for max_value, min_value, mean_value, median_value in zip(lh_ar_max, lh_ar_min, lh_ar_mean, lh_ar_median) ) )
dlg.recreate().set_current() dlg.recreate().set_current()

13
xlib/torch/optim/AdaBelief.py
View file

@ -1,5 +1,4 @@
import torch import torch
import numpy as np
class AdaBelief(torch.optim.Optimizer): class AdaBelief(torch.optim.Optimizer):
def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-16, def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-16,
@ -39,21 +38,15 @@ class AdaBelief(torch.optim.Optimizer):
grad.add_(p.data, alpha=group['weight_decay']) grad.add_(p.data, alpha=group['weight_decay'])
state['step'] += 1 state['step'] += 1
m_t, v_t = state['m_t'], state['v_t'] m_t, v_t = state['m_t'], state['v_t']
m_t.mul_(beta1).add_( grad , alpha=1 - beta1) m_t.mul_(beta1).add_( grad , alpha=1 - beta1)
v_t.mul_(beta2).add_( (grad - m_t)**2 , alpha=1 - beta2) v_t.mul_(beta2).add_( (grad - m_t)**2 , alpha=1 - beta2)
v_diff = (-group['lr'] * m_t).div_( v_t.sqrt().add_(group['eps']) ) v_diff = (-group['lr'] * m_t).div_( v_t.sqrt().add_(group['eps']) )
if group['lr_dropout'] < 1.0: if group['lr_dropout'] < 1.0:
lrd_rand = torch.ones_like(p.data) lrd_rand = torch.ones_like(p.data)
v_diff *= torch.bernoulli(lrd_rand * group['lr_dropout'] ) v_diff *= torch.bernoulli(lrd_rand * group['lr_dropout'] )
# from xlib.console.diacon import Diacon
# Diacon.stop()
# import code
# code.interact(local=dict(globals(), **locals()))
p.data.add_(v_diff) p.data.add_(v_diff)