add resnet v1 models, modify path settings.

data/scripts/fetch_coco_long_models.sh

@@ -4,6 +4,7 @@ DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )/../" && pwd )"
 cd $DIR
 
 FILE=coco_long.tgz
+# replace it with gs11655.sp.cs.cmu.edu if ladoga.graphics.cs.cmu.edu does not work
 URL=http://ladoga.graphics.cs.cmu.edu/xinleic/tf-faster-rcnn/$FILE
 CHECKSUM=099f637235d24ec97d9708b3ff66bd7f
 

data/scripts/fetch_faster_rcnn_models.sh

@@ -4,6 +4,7 @@ DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )/../" && pwd )"
 cd $DIR
 
 FILE=faster_rcnn_models.tgz
+# replace it with gs11655.sp.cs.cmu.edu if ladoga.graphics.cs.cmu.edu does not work
 URL=http://ladoga.graphics.cs.cmu.edu/xinleic/tf-faster-rcnn/$FILE
 CHECKSUM=865cdf7350a87ef41d6476e6e33b7212
 

data/scripts/fetch_imagenet_weights.sh

@@ -4,6 +4,7 @@ DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )/../" && pwd )"
 cd $DIR
 
 FILE=imagenet_weights.tgz
+# replace it with gs11655.sp.cs.cmu.edu if ladoga.graphics.cs.cmu.edu does not work
 URL=http://ladoga.graphics.cs.cmu.edu/xinleic/tf-faster-rcnn/$FILE
 CHECKSUM=e9772d7c761040f10a67d389336b90ce
 

experiments/cfgs/res50.yml

@@ -0,0 +1,17 @@
+EXP_DIR: res50
+TRAIN:
+  HAS_RPN: True
+  IMS_PER_BATCH: 1
+  BBOX_NORMALIZE_TARGETS_PRECOMPUTED: True
+  RPN_POSITIVE_OVERLAP: 0.7
+  RPN_BATCHSIZE: 256
+  PROPOSAL_METHOD: gt
+  BG_THRESH_LO: 0.0
+  DISPLAY: 20
+  BATCH_SIZE: 256
+  WEIGHT_DECAY: 0.0001
+  DOUBLE_BIAS: False
+  SNAPSHOT_PREFIX: res101_faster_rcnn
+TEST:
+  HAS_RPN: True
+POOLING_MODE: crop

lib/model/train_val.py

@@ -169,7 +169,10 @@ def train_model(self, sess, max_iters):
           var_to_dic[v.name] = v
           continue
         # exclude the first conv layer to swap RGB to BGR
-        if v.name == 'vgg_16/conv1/conv1_1/weights:0' or v.name == 'resnet_v1_101/conv1/weights:0':
+        if v.name == 'vgg_16/conv1/conv1_1/weights:0' 
+           or v.name == 'resnet_v1_50/conv1/weights:0'
+           or v.name == 'resnet_v1_101/conv1/weights:0'
+           or v.name == 'resnet_v1_152/conv1/weights:0':
           var_to_dic[v.name] = v
           continue
         if v.name.split(':')[0] in var_keep_dic:
@@ -200,16 +203,16 @@ def train_model(self, sess, max_iters):
                                 var_to_dic['vgg_16/fc7/weights:0'].get_shape())))
             sess.run(tf.assign(var_to_dic['vgg_16/conv1/conv1_1/weights:0'], 
                                 tf.reverse(conv1_rgb, [2])))
-      elif self.net._arch == 'res101':
-        print('Fix Resnet101 layers..')
-        with tf.variable_scope('Fix_Res101') as scope:
+      elif self.net._arch.startswith('res_v1'):
+        print('Fix Resnet V1 layers..')
+        with tf.variable_scope('Fix_Resnet_V1') as scope:
           with tf.device("/cpu:0"):
             # fix RGB to BGR
             conv1_rgb = tf.get_variable("conv1_rgb", [7, 7, 3, 64], trainable=False)
-            restorer_fc = tf.train.Saver({"resnet_v1_101/conv1/weights": conv1_rgb})
+            restorer_fc = tf.train.Saver({self.net._resnet_scope + "/conv1/weights": conv1_rgb})
             restorer_fc.restore(sess, self.pretrained_model)
 
-            sess.run(tf.assign(var_to_dic['resnet_v1_101/conv1/weights:0'], tf.reverse(conv1_rgb, [2])))
+            sess.run(tf.assign(var_to_dic[self.net._resnet_scope + '/conv1/weights:0'], tf.reverse(conv1_rgb, [2])))
       else:
         # every network should fix the rgb issue at least
         raise NotImplementedError

lib/nets/network.py

@@ -337,10 +337,11 @@ def create_architecture(self, sess, mode, num_classes, tag=None,
 
     return layers_to_output
 
+  # Extract the head feature maps, for example for vgg16 it is conv5_3
   # only useful during testing mode
-  def extract_conv5(self, sess, image):
+  def extract_head(self, sess, image):
     feed_dict = {self._image: image}
-    feat = sess.run(self._layers["conv5_3"], feed_dict=feed_dict)
+    feat = sess.run(self._layers["head"], feed_dict=feed_dict)
     return feat
 
   # only useful during testing mode

lib/nets/res101.py → lib/nets/resnet_v1.py

@@ -50,10 +50,12 @@ def resnet_arg_scope(is_training=True,
     with arg_scope([layers.batch_norm], **batch_norm_params) as arg_sc:
       return arg_sc
 
-class Resnet101(Network):
-  def __init__(self, batch_size=1):
+class resnetv1(Network):
+  def __init__(self, batch_size=1, num_layers=50):
     Network.__init__(self, batch_size=batch_size)
-    self._arch = 'res101'
+    self._num_layers = num_layers
+    self._arch = 'res_v1_%d' % num_layers
+    self._resnet_scope = 'resnet_v1_%d' % num_layers
 
   def _crop_pool_layer(self, bottom, rois, name):
     with tf.variable_scope(name) as scope:
@@ -81,7 +83,7 @@ def _crop_pool_layer(self, bottom, rois, name):
   # Do the first few layers manually, because 'SAME' padding can behave inconsistently
   # for images of different sizes: sometimes 0, sometimes 1
   def build_base(self):
-    with tf.variable_scope('resnet_v1_101', 'resnet_v1_101'):
+    with tf.variable_scope(self._resnet_scope, self._resnet_scope):
       net = resnet_utils.conv2d_same(self._image, 64, 7, stride=2, scope='conv1')
       net = tf.pad(net, [[0, 0], [1, 1], [1, 1], [0, 0]])
       net = slim.max_pool2d(net, [3, 3], stride=2, padding='VALID', scope='pool1')
@@ -97,57 +99,85 @@ def build_network(self, sess, is_training=True):
       initializer = tf.random_normal_initializer(mean=0.0, stddev=0.01)
       initializer_bbox = tf.random_normal_initializer(mean=0.0, stddev=0.001)
     bottleneck = resnet_v1.bottleneck
-    blocks = [
-      resnet_utils.Block('block1', bottleneck,
-                         [(256, 64, 1)] * 2 + [(256, 64, 2)]),
-      resnet_utils.Block('block2', bottleneck,
-                         [(512, 128, 1)] * 3 + [(512, 128, 2)]),
-      # Use stride-1 for the last conv4 layer
-      resnet_utils.Block('block3', bottleneck,
-                         [(1024, 256, 1)] * 22 + [(1024, 256, 1)]),
-      resnet_utils.Block('block4', bottleneck, [(2048, 512, 1)] * 3)
-    ]
+    # choose different blocks for different number of layers
+    if self._num_layers == 50:
+      blocks = [
+        resnet_utils.Block('block1', bottleneck,
+                           [(256, 64, 1)] * 2 + [(256, 64, 2)]),
+        resnet_utils.Block('block2', bottleneck,
+                           [(512, 128, 1)] * 3 + [(512, 128, 2)]),
+        # Use stride-1 for the last conv4 layer
+        resnet_utils.Block('block3', bottleneck,
+                           [(1024, 256, 1)] * 5 + [(1024, 256, 1)]),
+        resnet_utils.Block('block4', bottleneck, [(2048, 512, 1)] * 3)
+      ]
+    elif self._num_layers == 101:
+      blocks = [
+        resnet_utils.Block('block1', bottleneck,
+                           [(256, 64, 1)] * 2 + [(256, 64, 2)]),
+        resnet_utils.Block('block2', bottleneck,
+                           [(512, 128, 1)] * 3 + [(512, 128, 2)]),
+        # Use stride-1 for the last conv4 layer
+        resnet_utils.Block('block3', bottleneck,
+                           [(1024, 256, 1)] * 22 + [(1024, 256, 1)]),
+        resnet_utils.Block('block4', bottleneck, [(2048, 512, 1)] * 3)
+      ]
+    elif self._num_layers == 152:
+      blocks = [
+        resnet_utils.Block('block1', bottleneck,
+                           [(256, 64, 1)] * 2 + [(256, 64, 2)]),
+        resnet_utils.Block('block2', bottleneck,
+                           [(512, 128, 1)] * 7 + [(512, 128, 2)]),
+        # Use stride-1 for the last conv4 layer
+        resnet_utils.Block('block3', bottleneck,
+                           [(1024, 256, 1)] * 35 + [(1024, 256, 1)]),
+        resnet_utils.Block('block4', bottleneck, [(2048, 512, 1)] * 3)
+      ]
+    else:
+      # other numbers are not supported
+      raise NotImplementedError
+
     assert (0 <= cfg.RESNET.FIXED_BLOCKS < 4)
     if cfg.RESNET.FIXED_BLOCKS == 3:
       with slim.arg_scope(resnet_arg_scope(is_training=False)):
         net = self.build_base()
-        net_conv5, _ = resnet_v1.resnet_v1(net,
+        net_conv4, _ = resnet_v1.resnet_v1(net,
                                            blocks[0:cfg.RESNET.FIXED_BLOCKS],
                                            global_pool=False,
                                            include_root_block=False,
-                                           scope='resnet_v1_101')
+                                           scope=self._resnet_scope)
     elif cfg.RESNET.FIXED_BLOCKS > 0:
       with slim.arg_scope(resnet_arg_scope(is_training=False)):
         net = self.build_base()
         net, _ = resnet_v1.resnet_v1(net,
                                      blocks[0:cfg.RESNET.FIXED_BLOCKS],
                                      global_pool=False,
                                      include_root_block=False,
-                                     scope='resnet_v1_101')
+                                     scope=self._resnet_scope)
 
       with slim.arg_scope(resnet_arg_scope(is_training=is_training)):
-        net_conv5, _ = resnet_v1.resnet_v1(net,
+        net_conv4, _ = resnet_v1.resnet_v1(net,
                                            blocks[cfg.RESNET.FIXED_BLOCKS:-1],
                                            global_pool=False,
                                            include_root_block=False,
-                                           scope='resnet_v1_101')
+                                           scope=self._resnet_scope)
     else:  # cfg.RESNET.FIXED_BLOCKS == 0
       with slim.arg_scope(resnet_arg_scope(is_training=is_training)):
         net = self.build_base()
-        net_conv5, _ = resnet_v1.resnet_v1(net,
+        net_conv4, _ = resnet_v1.resnet_v1(net,
                                            blocks[0:-1],
                                            global_pool=False,
                                            include_root_block=False,
-                                           scope='resnet_v1_101')
+                                           scope=self._resnet_scope)
 
-    self._act_summaries.append(net_conv5)
-    self._layers['conv5_3'] = net_conv5
-    with tf.variable_scope('resnet_v1_101', 'resnet_v1_101'):
+    self._act_summaries.append(net_conv4)
+    self._layers['head'] = net_conv4
+    with tf.variable_scope(self._resnet_scope, self._resnet_scope):
       # build the anchors for the image
       self._anchor_component()
 
       # rpn
-      rpn = slim.conv2d(net_conv5, 512, [3, 3], trainable=is_training, weights_initializer=initializer,
+      rpn = slim.conv2d(net_conv4, 512, [3, 3], trainable=is_training, weights_initializer=initializer,
                         scope="rpn_conv/3x3")
       self._act_summaries.append(rpn)
       rpn_cls_score = slim.conv2d(rpn, self._num_anchors * 2, [1, 1], trainable=is_training,
@@ -176,7 +206,7 @@ def build_network(self, sess, is_training=True):
 
       # rcnn
       if cfg.POOLING_MODE == 'crop':
-        pool5 = self._crop_pool_layer(net_conv5, rois, "pool5")
+        pool5 = self._crop_pool_layer(net_conv4, rois, "pool5")
       else:
         raise NotImplementedError
 
@@ -185,9 +215,9 @@ def build_network(self, sess, is_training=True):
                                    blocks[-1:],
                                    global_pool=False,
                                    include_root_block=False,
-                                   scope='resnet_v1_101')
+                                   scope=self._resnet_scope)
 
-    with tf.variable_scope('resnet_v1_101', 'resnet_v1_101'):
+    with tf.variable_scope(self._resnet_scope, self._resnet_scope):
       # Average pooling done by reduce_mean
       fc7 = tf.reduce_mean(fc7, axis=[1, 2])
       cls_score = slim.fully_connected(fc7, self._num_classes, weights_initializer=initializer,

lib/nets/vgg16.py

@@ -46,7 +46,7 @@ def build_network(self, sess, is_training=True):
       net = slim.repeat(net, 3, slim.conv2d, 512, [3, 3],
                         trainable=is_training, scope='conv5')
       self._act_summaries.append(net)
-      self._layers['conv5_3'] = net
+      self._layers['head'] = net
       # build the anchors for the image
       self._anchor_component()
 

tools/_init_paths.py

@@ -11,4 +11,5 @@ def add_path(path):
 lib_path = osp.join(this_dir, '..', 'lib')
 add_path(lib_path)
 
-add_path('data/coco/PythonAPI')
+coco_path = osp.join(this_dir, '..', 'data', 'coco', 'PythonAPI')
+add_path(coco_path)

tools/demo.py

@@ -28,7 +28,7 @@
 import argparse
 
 from nets.vgg16 import vgg16
-from nets.res101 import Resnet101
+from nets.resnet_v1 import resnetv1
 
 CLASSES = ('__background__',
            'aeroplane', 'bicycle', 'bird', 'boat',
@@ -39,6 +39,7 @@
 
 NETS = {'vgg16': ('vgg16_faster_rcnn_iter_70000.ckpt',),'res101': ('res101_faster_rcnn_iter_110000.ckpt',)}
 DATASETS= {'pascal_voc': ('voc_2007_trainval',),'pascal_voc_0712': ('voc_2007_trainval+voc_2012_trainval',)}
+
 def vis_detections(im, class_name, dets, thresh=0.5):
     """Draw detected bounding boxes."""
     inds = np.where(dets[:, -1] >= thresh)[0]

tools/test_net.py

@@ -17,7 +17,7 @@
 
 import tensorflow as tf
 from nets.vgg16 import vgg16
-from nets.res101 import Resnet101
+from nets.resnet_v1 import resnetv1
 
 def parse_args():
   """
@@ -41,8 +41,8 @@ def parse_args():
                         help='tag of the model',
                         default='', type=str)
   parser.add_argument('--net', dest='net',
-                      help='vgg16 or res101',
-                      default='res101', type=str)
+                      help='vgg16, res50, res101, res152',
+                      default='res50', type=str)
   parser.add_argument('--set', dest='set_cfgs',
                         help='set config keys', default=None,
                         nargs=argparse.REMAINDER)
@@ -90,8 +90,12 @@ def parse_args():
   # load network
   if args.net == 'vgg16':
     net = vgg16(batch_size=1)
+  elif args.net == 'res50':
+    net = resnetv1(batch_size=1, num_layers=50)
   elif args.net == 'res101':
-    net = Resnet101(batch_size=1)
+    net = resnetv1(batch_size=1, num_layers=101)
+  elif args.net == 'res152':
+    net = resnetv1(batch_size=1, num_layers=152)
   else:
     raise NotImplementedError
 

tools/trainval_net.py

@@ -19,7 +19,7 @@
 
 import tensorflow as tf
 from nets.vgg16 import vgg16
-from nets.res101 import Resnet101
+from nets.resnet_v1 import resnetv1
 
 def parse_args():
   """
@@ -45,8 +45,8 @@ def parse_args():
                       help='tag of the model',
                       default=None, type=str)
   parser.add_argument('--net', dest='net',
-                      help='vgg16 or res101',
-                      default='res101', type=str)
+                      help='vgg16, res50, res101, res152',
+                      default='res50', type=str)
   parser.add_argument('--set', dest='set_cfgs',
                       help='set config keys', default=None,
                       nargs=argparse.REMAINDER)
@@ -119,10 +119,15 @@ def get_roidb(imdb_name):
   print('{:d} validation roidb entries'.format(len(valroidb)))
   cfg.TRAIN.USE_FLIPPED = orgflip
 
+  # load network
   if args.net == 'vgg16':
     net = vgg16(batch_size=cfg.TRAIN.IMS_PER_BATCH)
+  elif args.net == 'res50':
+    net = resnetv1(batch_size=cfg.TRAIN.IMS_PER_BATCH, num_layers=50)
   elif args.net == 'res101':
-    net = Resnet101(batch_size=cfg.TRAIN.IMS_PER_BATCH)
+    net = resnetv1(batch_size=cfg.TRAIN.IMS_PER_BATCH, num_layers=101)
+  elif args.net == 'res152':
+    net = resnetv1(batch_size=cfg.TRAIN.IMS_PER_BATCH, num_layers=152)
   else:
     raise NotImplementedError