Allow ignore_mismatch when loading a model to a session

examples/FasterRCNN/dataset/coco.py

@@ -235,6 +235,7 @@ def register_coco(basedir):
         DatasetRegistry.register(name, lambda x=split: COCODetection(basedir, x))
         DatasetRegistry.register_metadata(name, 'class_names', class_names)
 
+
 if __name__ == '__main__':
     basedir = '~/data/coco'
     c = COCODetection(basedir, 'train2014')

tensorpack/tfutils/sessinit.py

@@ -174,7 +174,9 @@ class SaverRestoreRelaxed(SaverRestore):
 
         def f(reader, name, v):
             val = reader.get_tensor(name)
-            v.load(SessionUpdate.relaxed_value_for_var(val, v))
+            val = SessionUpdate.relaxed_value_for_var(val, v, ignore_mismatch=True)
+            if val is not None:
+                v.load(val)
 
         with sess.as_default():
             self._match_vars(f)
@@ -185,14 +187,17 @@ class DictRestore(SessionInit):
     Restore variables from a dictionary.
     """
 
-    def __init__(self, variable_dict):
+    def __init__(self, variable_dict, ignore_mismatch=False):
         """
         Args:
             variable_dict (dict): a dict of {name: value}
+            ignore_mismatch (bool): ignore failures when the value and the
+                variable does not match.
         """
         assert isinstance(variable_dict, dict), type(variable_dict)
         # use varname (with :0) for consistency
         self._prms = {get_op_tensor_name(n)[1]: v for n, v in six.iteritems(variable_dict)}
+        self._ignore_mismatch = ignore_mismatch
 
     def _run_init(self, sess):
         variables = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
@@ -218,7 +223,7 @@ class DictRestore(SessionInit):
             mismatch.add(k)
         mismatch.log()
 
-        upd = SessionUpdate(sess, [v for v in variables if v.name in intersect])
+        upd = SessionUpdate(sess, [v for v in variables if v.name in intersect], ignore_mismatch=self._ignore_mismatch)
         logger.info("Restoring {} variables from dict ...".format(len(intersect)))
         upd.update({name: value for name, value in six.iteritems(self._prms) if name in intersect})
 
@@ -246,10 +251,15 @@ class ChainInit(SessionInit):
             i._run_init(sess)
 
 
-def get_model_loader(filename):
+def get_model_loader(filename, ignore_mismatch=False):
     """
     Get a corresponding model loader by looking at the file name.
 
+    Args:
+        filename (str): either a tensorflow checkpoint, or a npz file.
+        ignore_mismatch (bool): ignore failures when values in the file and
+            variables in the graph do not match.
+
     Returns:
         SessInit: either a :class:`DictRestore` (if name ends with 'npy/npz') or
         :class:`SaverRestore` (otherwise).
@@ -258,10 +268,13 @@ def get_model_loader(filename):
     filename = os.path.expanduser(filename)
     if filename.endswith('.npy'):
         assert tf.gfile.Exists(filename), filename
-        return DictRestore(np.load(filename, encoding='latin1').item())
+        return DictRestore(np.load(filename, encoding='latin1').item(), ignore_mismatch=ignore_mismatch)
     elif filename.endswith('.npz'):
         assert tf.gfile.Exists(filename), filename
         obj = np.load(filename)
-        return DictRestore(dict(obj))
+        return DictRestore(dict(obj), ignore_mismatch=ignore_mismatch)
     else:
-        return SaverRestore(filename)
+        if ignore_mismatch:
+            return SaverRestoreRelaxed(filename)
+        else:
+            return SaverRestore(filename)

tensorpack/tfutils/varmanip.py

@@ -38,17 +38,20 @@ def get_savename_from_varname(
 class SessionUpdate(object):
     """ Update the variables in a session """
 
-    def __init__(self, sess, vars_to_update):
+    def __init__(self, sess, vars_to_update, ignore_mismatch=False):
         """
         Args:
             sess (tf.Session): a session object
             vars_to_update: a collection of variables to update
+            ignore_mismatch (bool): ignore failures when the value and the
+                variable does not match.
         """
         self.sess = sess
         self.name_map = {v.name: v for v in vars_to_update}
+        self.ignore_mismatch = ignore_mismatch
 
     @staticmethod
-    def relaxed_value_for_var(value, var):
+    def relaxed_value_for_var(value, var, ignore_mismatch=False):
         """
         Returns a relaxed (possibly reshaped/upcast-ed) version of value,
         to be loaded to the given variable.
@@ -56,9 +59,13 @@ class SessionUpdate(object):
         Args:
             value (ndarray): an numpy array to be loaded to var
             var (tf.Variable):
+            ignore_mismatch (bool): ignore failures when the value and the
+                variable does not match.
 
         Returns:
-            ndarray: a possibly reshaped or casted version of value
+            ndarray: a possibly reshaped or casted version of value.
+            Returns None if `ignore_mismatch==True` and the value and the variable
+            mismatch.
         """
         assert isinstance(var, tf.Variable)
         name = var.op.name
@@ -66,11 +73,17 @@ class SessionUpdate(object):
         # check incompatible shape
         varshape = tuple(var.get_shape().as_list())
         if varshape != value.shape:
-            # TODO only allow reshape when shape different by empty axis
             if np.prod(varshape) != np.prod(value.shape):
-                raise ValueError(
-                    "Trying to load a tensor of shape {} into the variable '{}' whose shape is {}.".format(
-                        value.shape, name, varshape))
+                if ignore_mismatch:
+                    logger.warn(
+                        "Cannot load a tensor of shape {} into the variable '{}' whose shape is {}.".format(
+                            value.shape, name, varshape))
+                    return None
+                else:
+                    raise ValueError(
+                        "Trying to load a tensor of shape {} into the variable '{}' whose shape is {}.".format(
+                            value.shape, name, varshape))
+            # TODO only allow reshape when shape different by empty axis
             logger.warn("The tensor is reshaped from {} to {} when assigned to '{}'".format(
                 value.shape, varshape, name))
             value = value.reshape(varshape)
@@ -115,9 +128,12 @@ class SessionUpdate(object):
             for name, value in six.iteritems(prms):
                 assert name in self.name_map
                 var = self.name_map[name]
-                fetches.append(var.initializer)
+                value = SessionUpdate.relaxed_value_for_var(
+                    value, var, ignore_mismatch=self.ignore_mismatch)
                 # This is the implementation of `var.load`
-                feeds[var.initializer.inputs[1]] = SessionUpdate.relaxed_value_for_var(value, var)
+                if value is not None:
+                    fetches.append(var.initializer)
+                    feeds[var.initializer.inputs[1]] = value
             self.sess.run(fetches, feed_dict=feeds)