[MaskRCNN] use small buffer size in data loader to reduce memory usage (#1164,#1152,#1111)

examples/FasterRCNN/config.py

@@ -93,7 +93,7 @@ _C.DATA.NUM_CATEGORY = 0  # without the background class (e.g., 80 for COCO)
 _C.DATA.CLASS_NAMES = []  # NUM_CLASS (NUM_CATEGORY+1) strings, the first is "BG".
 # whether the coordinates in the annotations are absolute pixel values, or a relative value in [0, 1]
 _C.DATA.ABSOLUTE_COORD = True
-_C.DATA.NUM_WORKERS = 5  # number of data loading workers
+_C.DATA.NUM_WORKERS = 5  # number of data loading workers. set to 0 to disable parallel data loading
 
 # backbone ----------------------
 _C.BACKBONE.WEIGHTS = ''   # /path/to/weights.npz

examples/FasterRCNN/data.py

@@ -8,7 +8,7 @@ from tabulate import tabulate
 from termcolor import colored
 
 from tensorpack.dataflow import (
-    DataFromList, MapDataComponent, MultiProcessMapDataZMQ, MultiThreadMapData, TestDataSpeed, imgaug)
+    DataFromList, MapDataComponent, MapData, MultiProcessMapDataZMQ, MultiThreadMapData, TestDataSpeed, imgaug)
 from tensorpack.utils import logger
 from tensorpack.utils.argtools import log_once, memoized
 
@@ -368,11 +368,15 @@ def get_train_dataflow():
             # tpviz.interactive_imshow(viz)
         return ret
 
-    if cfg.TRAINER == 'horovod':
-        ds = MultiThreadMapData(ds, cfg.DATA.NUM_WORKERS, preprocess)
-        # MPI does not like fork()
+    if cfg.DATA.NUM_WORKERS > 0:
+        buffer_size = cfg.DATA.NUM_WORKERS * 20
+        if cfg.TRAINER == 'horovod':
+            ds = MultiThreadMapData(ds, cfg.DATA.NUM_WORKERS, preprocess, buffer_size=buffer_size)
+            # MPI does not like fork()
+        else:
+            ds = MultiProcessMapDataZMQ(ds, cfg.DATA.NUM_WORKERS, preprocess, buffer_size=buffer_size)
     else:
-        ds = MultiProcessMapDataZMQ(ds, cfg.DATA.NUM_WORKERS, preprocess)
+        ds = MapData(ds, preprocess)
     return ds
 
 

tensorpack/dataflow/parallel_map.py

@@ -154,6 +154,7 @@ class MultiThreadMapData(_ParallelMapData):
             strict (bool): use "strict mode", see notes above.
         """
         super(MultiThreadMapData, self).__init__(ds, buffer_size, strict)
+        assert nr_thread > 0, nr_thread
 
         self._strict = strict
         self.nr_thread = nr_thread
@@ -259,6 +260,7 @@ class MultiProcessMapDataZMQ(_ParallelMapData, _MultiProcessZMQDataFlow):
         """
         _ParallelMapData.__init__(self, ds, buffer_size, strict)
         _MultiProcessZMQDataFlow.__init__(self)
+        assert nr_proc > 0, nr_proc
         self.nr_proc = nr_proc
         self.map_func = map_func
         self._strict = strict