tiny

src/age_iter.py

@@ -163,6 +163,8 @@ class FaceImageIter(io.DataIter):
         try:
             while i < batch_size:
                 label, s, bbox, landmark = self.next_sample()
+                #if label[1]>=0.0 or label[2]>=0.0:
+                #  print(label[0:10])
                 _data = self.imdecode(s)
                 if self.rand_mirror:
                   _rd = random.randint(0,1)

src/data/age_merge.py

@@ -24,6 +24,10 @@ def main(args):
     stat = [0,0]
     for ds in ['ms1m', 'megaage', 'imdb']:
         for n in ['train', 'val']:
+            if ds=='ms1m' or ds=='imdb':
+              continue
+            if n=='val' and ds!='megaage':
+              continue
             repeat = 1
             if n=='train' and ds=='megaage':
                 repeat = 1
@@ -34,7 +38,7 @@ def main(args):
             if n=='val':
                 writer = val_writer
                 widx = val_widx
-            path = os.path.join(ds, '%s.rec'%n)
+            path = os.path.join(args.input, ds, '%s.rec'%n)
             if not os.path.exists(path):
                 continue
             imgrec = mx.recordio.MXIndexedRecordIO(path[:-3]+'idx', path, 'r')  # pylint: disable=redefined-variable-type
@@ -93,6 +97,7 @@ def main(args):
 if __name__ == '__main__':
   parser = argparse.ArgumentParser(description='do dataset merge')
   # general
+  parser.add_argument('--input', default='', type=str, help='')
   parser.add_argument('--output', default='', type=str, help='')
   args = parser.parse_args()
   main(args)

src/train_age.py

@@ -50,17 +50,21 @@ USE_AGE = True
 
 
 class AccMetric(mx.metric.EvalMetric):
-  def __init__(self, pred_idx = 1, name='acc'):
+  def __init__(self, pred_idx = 1, label_idx = 0, name='acc'):
     self.axis = 1
     self.pred_idx = pred_idx
+    self.label_idx = label_idx
     super(AccMetric, self).__init__(
-        'name', axis=self.axis,
+        name, axis=self.axis,
         output_names=None, label_names=None)
+    self.name = name
     self.losses = []
     self.count = 0
 
   def update(self, labels, preds):
     self.count+=1
+
+    #print('label num', len(labels))
     preds = [preds[self.pred_idx]] #use softmax output
     for label, pred_label in zip(labels, preds):
         if pred_label.shape != label.shape:
@@ -68,8 +72,9 @@ class AccMetric(mx.metric.EvalMetric):
         pred_label = pred_label.asnumpy().astype('int32').flatten()
         label = label.asnumpy()
         if label.ndim==2:
-          label = label[:,0]
+          label = label[:,self.label_idx]
         label = label.astype('int32').flatten()
+        print(self.name, label)
         assert label.shape==pred_label.shape
         self.sum_metric += (pred_label.flat == label.flat).sum()
         self.num_inst += len(pred_label.flat)
@@ -331,14 +336,17 @@ def train_net(args):
     data_dir = data_dir_list[0]
     path_imgrec = None
     path_imglist = None
-    prop = face_image.load_property(data_dir)
-    args.num_classes = prop.num_classes
-    image_size = prop.image_size
+    args.num_classes = 0
+    image_size = (112,112)
+    if os.path.exists(os.path.join(data_dir, 'property')):
+      prop = face_image.load_property(data_dir)
+      args.num_classes = prop.num_classes
+      image_size = prop.image_size
+      assert(args.num_classes>0)
+      print('num_classes', args.num_classes)
     args.image_h = image_size[0]
     args.image_w = image_size[1]
     print('image_size', image_size)
-    assert(args.num_classes>0)
-    print('num_classes', args.num_classes)
     path_imgrec = os.path.join(data_dir, "train.rec")
 
     print('Called with argument:', args)
@@ -392,13 +400,13 @@ def train_net(args):
 
     eval_metrics = []
     if USE_FR:
-      _metric = AccMetric(pred_idx=1)
+      _metric = AccMetric(pred_idx=1, label_idx=0)
       eval_metrics.append(_metric)
       if USE_GENDER:
-          _metric = AccMetric(pred_idx=2, name='gender')
+          _metric = AccMetric(pred_idx=2, label_idx=1, name='gender')
           eval_metrics.append(_metric)
     elif USE_GENDER:
-      _metric = AccMetric(pred_idx=1, name='gender')
+      _metric = AccMetric(pred_idx=1, label_idx=1, name='gender')
       eval_metrics.append(_metric)
     if USE_AGE:
       _metric = MAEMetric()