update

jfda/lnet.py

@@ -11,7 +11,7 @@
 import caffe
 import numpy as np
 from caffe.proto import caffe_pb2
-import google.protobuf as pb2
+from google.protobuf import text_format
 from jfda.config import cfg
 from jfda.utils import load_celeba, get_logger, crop_face
 
@@ -191,7 +191,7 @@ def get_data_size(txt):
   final_model = 'tmp/lnet_iter_%d.caffemodel'%max_iter
   solver_param = caffe_pb2.SolverParameter()
   with open('proto/l_solver.prototxt', 'r') as fin:
-    pb2.text_format.Merge(fin.read(), solver_param)
+    text_format.Merge(fin.read(), solver_param)
   solver_param.max_iter = max_iter
   solver_param.snapshot = iter_train
   solver_param.test_interval = iter_train
@@ -201,7 +201,7 @@ def get_data_size(txt):
   solver_param.stepsize = args.lrp * iter_train
   tmp_solver_prototxt = 'tmp/l_solver.prototxt'
   with open(tmp_solver_prototxt, 'w') as fout:
-    fout.write(pb2.text_format.MessageToString(solver_param))
+    fout.write(text_format.MessageToString(solver_param))
   # solver setup
   solver = caffe.SGDSolver(tmp_solver_prototxt)
   # train

jfda/train.py

@@ -7,7 +7,7 @@
 import numpy as np
 import caffe
 from caffe.proto import caffe_pb2
-import google.protobuf as pb2
+from google.protobuf import text_format
 from jfda.config import cfg
 from jfda.minibatch import MiniBatcher
 
@@ -45,7 +45,7 @@ def __init__(self, solver_prototxt, args):
     self.final_model = 'tmp/%snet_iter_%d.caffemodel'%(net_type, max_iter)
     solver_param = caffe_pb2.SolverParameter()
     with open(solver_prototxt, 'r') as fin:
-      pb2.text_format.Merge(fin.read(), solver_param)
+      text_format.Merge(fin.read(), solver_param)
     solver_param.max_iter = max_iter  # max training iterations
     solver_param.snapshot = iter_train  # save after an epoch
     solver_param.test_interval = iter_train
@@ -56,7 +56,7 @@ def __init__(self, solver_prototxt, args):
     solver_param.weight_decay = args.wd
     tmp_solver_prototxt = 'tmp/%s_solver.prototxt'%net_type
     with open(tmp_solver_prototxt, 'w') as fout:
-      fout.write(pb2.text_format.MessageToString(solver_param))
+      fout.write(text_format.MessageToString(solver_param))
     # solver setup
     self.solver = caffe.SGDSolver(tmp_solver_prototxt)
     # data layer setup

mxnet/model.py

@@ -97,5 +97,4 @@ def lnet():
     mx.viz.plot_network(p, shape={'data': (1, 3, 12, 12)}).render('tmp/pnet')
     mx.viz.plot_network(r, shape={'data': (1, 3, 24, 24)}).render('tmp/rnet')
     mx.viz.plot_network(o, shape={'data': (1, 3, 48, 48)}).render('tmp/onet')
-    # mx.viz doesn't support multi-output from an op
-    #mx.viz.plot_network(l, shape={'data': (1, 15, 24, 24)}).render('tmp/lnet')
+    mx.viz.plot_network(l, shape={'data': (1, 15, 24, 24)}).render('tmp/lnet')

train.sh

@@ -1,27 +1,37 @@
 #!/usr/bin/env bash
 
+set -e
 GPU=0
+GL=$GLOG_minloglevel
 
 # pnet
 echo "Generate Data for pNet"
+export GLOG_minloglevel=2
 python jfda/prepare.py --net p --wider --celeba --worker 8
 echo "Train pNet"
+export GLOG_minloglevel=$GL
 python jfda/train.py --net p --gpu $GPU --size 128 --lr 0.05 --lrw 0.1 --lrp 5 --wd 0.0001 --epoch 25
 
 # rnet
 echo "Generate Data for rNet"
+export GLOG_minloglevel=2
 python jfda/prepare.py --net r --gpu $GPU --detect --celeba --wider --worker 4
 echo "Train rNet"
+export GLOG_minloglevel=$GL
 python jfda/train.py --net r --gpu $GPU --size 128 --lr 0.05 --lrw 0.1 --lrp 5 --wd 0.0001 --epoch 25
 
 # onet
 echo "Generate Data for oNet"
+export GLOG_minloglevel=2
 python jfda/prepare.py --net o --gpu $GPU --detect --celeba --wider --worker 4
 echo "Train oNet"
+export GLOG_minloglevel=$GL
 python jfda/train.py --net o --gpu $GPU --size 64 --lr 0.05 --lrw 0.1 --lrp 7 --wd 0.0001 --epoch 35
 
 # lnet
 echo "Generate Data for lNet"
+export GLOG_minloglevel=2
 python jfda/lnet.py --prepare --worker 8
 echo "Train lNet"
+export GLOG_minloglevel=$GL
 python jfda/lnet.py --train --gpu $GPU --lr 0.1 --lrw 0.1 --lrp 2 --epoch 10