fix some bugs

tensorpack/dataflow/image.py

@@ -16,7 +16,7 @@ class ImageFromFile(DataFlow):
             channel: 1 or 3 channel
             resize: a (h, w) tuple. If given, will force a resize
         """
-        assert len(self.files)
+        assert len(files)
         self.files = files
         self.channel = int(channel)
         self.resize = resize

tensorpack/models/pool.py

@@ -39,7 +39,7 @@ def FixedUnPooling(x, shape, unpool_mat=None):
     input_shape = x.get_shape().as_list()
     assert len(input_shape) == 4
     if unpool_mat is None:
-        mat = np.zeros(shape)
+        mat = np.zeros(shape, dtype='float32')
         mat[0][0] = 1
         unpool_mat = tf.Variable(mat, trainable=False, name='unpool_mat')
     assert unpool_mat.get_shape().as_list() == list(shape)

tensorpack/train.py

@@ -103,6 +103,8 @@ def start_train(config):
 
     def get_model_inputs():
         model_inputs = input_queue.dequeue()
+        if isinstance(model_inputs, tf.Tensor):
+            model_inputs = [model_inputs]
         for qv, v in zip(model_inputs, input_vars):
             if config.batched_model_input:
                 qv.set_shape(v.get_shape())
@@ -179,7 +181,7 @@ def start_train(config):
             raise
         finally:
             coord.request_stop()
-            queue.close(cancel_pending_enqueues=True)
+            input_queue.close(cancel_pending_enqueues=True)
             callbacks.after_train()
             sess.close()
 

tensorpack/utils/logger.py

@@ -55,8 +55,9 @@ def set_file(path):
 global LOG_FILE
 LOG_FILE = "train_log/log.log"
 def set_logger_file(filename):
-    global LOG_FILE
+    global LOG_FILE, LOG_DIR
     LOG_FILE = filename
+    LOG_DIR = os.path.dirname(LOG_FILE)
     mkdir_p(os.path.dirname(LOG_FILE))
     set_file(LOG_FILE)
 

tensorpack/utils/symbolic_functions.py

@@ -18,8 +18,7 @@ def one_hot(y, num_labels):
         return tf.cast(onehot_labels, tf.float32)
 
 def flatten(x):
-    total_dim = np.prod(x.get_shape().as_list())
-    return tf.reshape(x, [total_dim])
+    return tf.reshape(x, [-1])
 
 def batch_flatten(x):
     total_dim = np.prod(x.get_shape()[1:].as_list())