def load(self, ckpt_path):
load_dict = {
'detector': self.detector,
}
if opt.resume:
load_dict.update({
'optimizer': self.optimizer,
'scheduler': self.scheduler,
})
utils.color_print(
'Load checkpoint from %s, resume training.' % ckpt_path, 3)
else:
utils.color_print('Load checkpoint from %s.' % ckpt_path, 3)
ckpt_info = load_checkpoint(load_dict,
ckpt_path,
map_location=opt.device)
s = torch.load(ckpt_path)
if opt.resume:
self.optimizer.load_state_dict(s['optimizer'])
epoch = ckpt_info.get('epoch', 0)
return epoch
def load(self, ckpt_path):
if ckpt_path[-2:] != 'pt':
return 0
load_dict = {
'detector': self.detector,
}
if opt.resume or 'RESUME' in self.config.MISC:
load_dict.update({
'optimizer': self.optimizer,
'scheduler': self.scheduler,
})
utils.color_print('Load checkpoint from %s, resume training.' % ckpt_path, 3)
else:
utils.color_print('Load checkpoint from %s.' % ckpt_path, 3)
ckpt_info = load_checkpoint(load_dict, ckpt_path, map_location=opt.device)
s = torch.load(ckpt_path, map_location='cpu')
if opt.resume or 'RESUME' in self.config.MISC:
self.optimizer.load_state_dict(s['optimizer'])
self.scheduler.step()
epoch = ckpt_info.get('epoch', 0)
return epoch
def load(self, ckpt_path):
load_dict = {
'cleaner': self.cleaner,
}
if opt.resume:
load_dict.update({
'optimizer': self.g_optimizer,
'scheduler': self.scheduler,
})
utils.color_print('Load checkpoint from %s, resume training.' % ckpt_path, 3)
else:
utils.color_print('Load checkpoint from %s.' % ckpt_path, 3)
ckpt_info = load_checkpoint(load_dict, ckpt_path, map_location=opt.device)
epoch = ckpt_info.get('epoch', 0)
return epoch