fix equals

Author: HiKapok

eval_ron_network.py

@@ -62,9 +62,9 @@
 #     'match_threshold', 0.5, 'Matching threshold with groundtruth objects.')
 
 tf.app.flags.DEFINE_float(
-    'select_threshold', 0.51, 'Selection threshold.')
+    'select_threshold', 0.01, 'Selection threshold.')
 tf.app.flags.DEFINE_float(
-    'objectness_thres', 0.93, 'threshold for the objectness to indicate the exist of object in that location.')
+    'objectness_thres', 0.03, 'threshold for the objectness to indicate the exist of object in that location.')
 tf.app.flags.DEFINE_integer(
     'select_top_k', 200, 'Select top-k detected bounding boxes.')
 tf.app.flags.DEFINE_integer(
@@ -100,7 +100,7 @@
 tf.app.flags.DEFINE_string(
     'master', '', 'The address of the TensorFlow master to use.')
 tf.app.flags.DEFINE_string(
-    'checkpoint_path', './model/model.ckpt-60399',#118815
+    'checkpoint_path', './model/model.ckpt-122044',#118815
     'The directory where the model was written to or an absolute path to a '
     'checkpoint file.')
 tf.app.flags.DEFINE_string(

nets/ron_vgg_320.py

@@ -120,7 +120,7 @@ class RONNet(object):
         # anchor_steps=[64],
 
         anchor_offset=0.5,
-        prior_scaling=[0.1, 0.1, 0.2, 0.2]
+        prior_scaling=[1., 1., 1., 1.]#[0.1, 0.1, 0.2, 0.2]
         )
 
     def __init__(self, params=None):
@@ -608,7 +608,7 @@ def ron_arg_scope(weight_decay=0.0005, is_training=True, data_format='NHWC'):
                         biases_initializer=tf.zeros_initializer()):
       with slim.arg_scope([slim.conv2d], activation_fn=tf.nn.relu,
                         weights_regularizer=slim.l2_regularizer(weight_decay),
-                        weights_initializer=truncated_normal_001_initializer(),
+                        weights_initializer=tf.contrib.layers.xavier_initializer(),#truncated_normal_001_initializer(),
                         biases_initializer=tf.zeros_initializer()):
         with slim.arg_scope([slim.conv2d, slim.conv2d_transpose, slim.max_pool2d],
                             padding='SAME',
@@ -691,7 +691,7 @@ def ron_losses(logits, localisations, objness_logits, objness_pred,
         # negtive examples are those max_overlap is still lower than neg_threshold, note that some positive may also has lower jaccard
 
         #negtive_mask = tf.cast(tf.logical_not(positive_mask), dtype) * gscores < neg_threshold
-        negtive_mask = (gclasses == 0)
+        negtive_mask = tf.equal(gclasses, 0) #(gclasses == 0)
         #negtive_mask = tf.logical_and(gscores < neg_threshold, tf.logical_not(positive_mask))
         fnegtive_mask = tf.cast(negtive_mask, dtype)
         n_negtives = tf.reduce_sum(fnegtive_mask)
@@ -769,7 +769,7 @@ def ron_losses(logits, localisations, objness_logits, objness_pred,
             loss = custom_layers.modified_smooth_l1(localisations, tf.stop_gradient(glocalisations), sigma = 3.)
             #loss = custom_layers.abs_smooth(localisations - tf.stop_gradient(glocalisations))
 
-            loss = tf.cond(n_cls_positives > 0., lambda: beta * tf.reduce_mean(tf.boolean_mask(tf.reduce_sum(loss, axis=-1), tf.stop_gradient(cls_positive_mask))), lambda: 0.)
+            loss = tf.cond(n_cls_positives > 0., lambda: beta * n_cls_positives / total_examples_for_cls * tf.reduce_mean(tf.boolean_mask(tf.reduce_sum(loss, axis=-1), tf.stop_gradient(cls_positive_mask))), lambda: 0.)
             #loss = tf.cond(n_positives > 0., lambda: beta * n_positives / total_examples_for_objness * tf.reduce_mean(tf.boolean_mask(tf.reduce_sum(loss, axis=-1), tf.stop_gradient(positive_mask))), lambda: 0.)
             #loss = tf.reduce_mean(loss * weights)
             #loss = tf.reduce_sum(loss * weights)

nets/ssd_common.py

@@ -46,7 +46,7 @@ def iou_matrix(bboxes, gt_bboxes):
     return tf.where(tf.equal(inter_vol, 0.0),
                     tf.zeros_like(inter_vol), tf.truediv(inter_vol, union_vol))
 
-def do_dual_max_match(overlap_matrix, high_thres, low_thres, ignore_between = True, gt_max_first=False):
+def do_dual_max_match(overlap_matrix, high_thres, low_thres, ignore_between = True, gt_max_first=True):
     '''
     overlap_matrix: num_gt * num_anchors
     '''
@@ -67,7 +67,7 @@ def do_dual_max_match(overlap_matrix, high_thres, low_thres, ignore_between = Tr
     gt_to_anchors = tf.argmax(overlap_matrix, axis=1)
 
     if gt_max_first:
-        left_gt_to_anchors_mask = tf.one_hot(gt_to_anchors, tf.shape(overlap_matrix)[1], on_value=1, off_value=0, axis=1, dtype=tf.int64)
+        left_gt_to_anchors_mask = tf.one_hot(gt_to_anchors, tf.shape(overlap_matrix)[1], on_value=1, off_value=0, axis=1, dtype=tf.int32)
     else:
         left_gt_to_anchors_mask = tf.cast(tf.logical_and(tf.reduce_max(anchors_to_gt_mask, axis=1, keep_dims=True) < 1, tf.one_hot(gt_to_anchors, tf.shape(overlap_matrix)[1], on_value=True, off_value=False, axis=1, dtype=tf.bool)), tf.int64)
 

ron_net.py

@@ -44,9 +44,9 @@
 # =========================================================================== #
 
 tf.app.flags.DEFINE_float(
-    'loss_alpha', 2./5, 'Alpha parameter in the loss function.')
+    'loss_alpha', 1./3, 'Alpha parameter in the loss function.')
 tf.app.flags.DEFINE_float(
-    'loss_beta', 2./5, 'Beta parameter in the loss function.')
+    'loss_beta', 1./3, 'Beta parameter in the loss function.')
 tf.app.flags.DEFINE_float(
     'negative_ratio', 3., 'Negative ratio in the loss function.')
 tf.app.flags.DEFINE_float(
@@ -154,7 +154,7 @@
 # Fine-Tuning Flags.
 # =========================================================================== #
 tf.app.flags.DEFINE_string(
-    'checkpoint_path', '../vgg_model/reduced/vgg16_reducedfc.ckpt',#None, #'../vgg_model/reduced/vgg16_reducedfc.ckpt'
+    'checkpoint_path', None,#None, #'../vgg_model/reduced/vgg16_reducedfc.ckpt'
     'The path to a checkpoint from which to fine-tune.')
 tf.app.flags.DEFINE_string(
     'checkpoint_model_scope', 'vgg_16',#None,
@@ -359,7 +359,7 @@ def wrapper_debug(sess):
                 logdir=FLAGS.model_dir,
                 master='',
                 is_chief=True,
-                init_fn=tf_utils.get_init_fn(FLAGS, os.path.join(FLAGS.data_dir, 'vgg_16.ckpt')),
+                init_fn=tf_utils.get_init_fn(FLAGS, os.path.join(FLAGS.data_dir, 'vgg_model/vgg16_reducedfc.ckpt')),
                 summary_op=summary_op,
                 number_of_steps=FLAGS.max_number_of_steps,
                 log_every_n_steps=FLAGS.log_every_n_steps,