deepinsight / deepinsight/insightface
A question about retinaface training code.
- Dominant language
- Python
- Stars
- 29.7k
- Forks
- 6.1k
- PR merge metrics
- No merged PRs in 30d
Description
I found that mnet has some fix params, "stage1*" and "upsampling" and layers whose index less than 7,such as
**def get_fixed_params(symbol, fixed_param):
if not config.LAYER_FIX:
return []
fixed_param_names = []
#for name in symbol.list_arguments():
# for f in fixed_param:
# if re.match(f, name):
# fixed_param_names.append(name)
#pre = 'mobilenetv20_features_linearbottleneck'
idx = 0
for name in symbol.list_arguments():
#print(idx, name)
if idx < 7 and name != 'data':
fixed_param_names.append(name)
#elif name.startswith('stage1_'):
# fixed_param_names.append(name)
if name.find('upsampling') >= 0:
fixed_param_names.append(name)
idx += 1
return fixed_param_names**
I don't konw why.
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.