Merge pull request #523 from Hakuyume/improve-apply-prediction

Improve apply_prediction

chainercv/extensions/evaluator/detection_voc_evaluator.py

@@ -5,7 +5,7 @@
 import chainer.training.extensions
 
 from chainercv.evaluations import eval_detection_voc
-from chainercv.utils import apply_prediction_to_iterator
+from chainercv.utils import apply_to_iterator
 
 
 class DetectionVOCEvaluator(chainer.training.extensions.Evaluator):
@@ -72,17 +72,17 @@ def evaluate(self):
         else:
             it = copy.copy(iterator)
 
-        imgs, pred_values, gt_values = apply_prediction_to_iterator(
+        in_values, out_values, rest_values = apply_to_iterator(
             target.predict, it)
-        # delete unused iterator explicitly
-        del imgs
+        # delete unused iterators explicitly
+        del in_values
 
-        pred_bboxes, pred_labels, pred_scores = pred_values
+        pred_bboxes, pred_labels, pred_scores = out_values
 
-        if len(gt_values) == 3:
-            gt_bboxes, gt_labels, gt_difficults = gt_values
-        elif len(gt_values) == 2:
-            gt_bboxes, gt_labels = gt_values
+        if len(rest_values) == 3:
+            gt_bboxes, gt_labels, gt_difficults = rest_values
+        elif len(rest_values) == 2:
+            gt_bboxes, gt_labels = rest_values
             gt_difficults = None
 
         result = eval_detection_voc(

chainercv/extensions/evaluator/semantic_segmentation_evaluator.py

@@ -5,7 +5,7 @@
 import chainer.training.extensions
 
 from chainercv.evaluations import eval_semantic_segmentation
-from chainercv.utils import apply_prediction_to_iterator
+from chainercv.utils import apply_to_iterator
 
 
 class SemanticSegmentationEvaluator(chainer.training.extensions.Evaluator):
@@ -79,13 +79,13 @@ def evaluate(self):
         else:
             it = copy.copy(iterator)
 
-        imgs, pred_values, gt_values = apply_prediction_to_iterator(
+        in_values, out_values, rest_values = apply_to_iterator(
             target.predict, it)
-        # delete unused iterator explicitly
-        del imgs
+        # delete unused iterators explicitly
+        del in_values
 
-        pred_labels, = pred_values
-        gt_labels, = gt_values
+        pred_labels, = out_values
+        gt_labels, = rest_values
 
         result = eval_semantic_segmentation(pred_labels, gt_labels)
 

chainercv/utils/__init__.py

@@ -6,7 +6,7 @@
 from chainercv.utils.image import read_image  # NOQA
 from chainercv.utils.image import tile_images  # NOQA
 from chainercv.utils.image import write_image  # NOQA
-from chainercv.utils.iterator import apply_prediction_to_iterator  # NOQA
+from chainercv.utils.iterator import apply_to_iterator  # NOQA
 from chainercv.utils.iterator import ProgressHook  # NOQA
 from chainercv.utils.iterator import unzip  # NOQA
 from chainercv.utils.testing import assert_is_bbox  # NOQA

chainercv/utils/iterator/__init__.py

@@ -1,3 +1,3 @@
-from chainercv.utils.iterator.apply_prediction_to_iterator import apply_prediction_to_iterator  # NOQA
+from chainercv.utils.iterator.apply_to_iterator import apply_to_iterator  # NOQA
 from chainercv.utils.iterator.progress_hook import ProgressHook  # NOQA
 from chainercv.utils.iterator.unzip import unzip  # NOQA

chainercv/utils/iterator/apply_to_iterator.py

@@ -0,0 +1,169 @@
+from chainercv.utils.iterator.unzip import unzip
+
+
+def apply_to_iterator(func, iterator, n_input=1, hook=None):
+    """Apply a function/method to batches from an iterator.
+
+    This function applies a function/method to an iterator of batches.
+
+    It assumes that the iterator iterates over a collection of tuples
+    that contain inputs to :func:`func`.
+    Additionally, the tuples may contain values
+    that are not used by :func:`func`.
+    For convenience, we allow the iterator to iterate over a collection of
+    inputs that are not tuple.
+    Here is an illustration of the expected behavior of the iterator.
+    This behaviour is the same as :class:`chainer.Iterator`.
+
+    >>> batch = next(iterator)
+    >>> # batch: [in_val]
+    or
+    >>> # batch: [(in_val0, ..., in_val{n_input - 1})]
+    or
+    >>> # batch: [(in_val0, ..., in_val{n_input - 1}, rest_val0, ...)]
+
+    :func:`func` should take batch(es) of data and
+    return batch(es) of computed values.
+    Here is an illustration of the expected behavior of the function.
+
+    >>> out_vals = func([in_val0], ..., [in_val{n_input - 1}])
+    >>> # out_vals: [out_val]
+    or
+    >>> out_vals0, out_vals1, ... = func([in_val0], ..., [in_val{n_input - 1}])
+    >>> # out_vals0: [out_val0]
+    >>> # out_vals1: [out_val1]
+
+    With :func:`apply_to_iterator`, users can get iterator(s) of values
+    returned by :func:`func`. It also returns iterator(s) of input values and
+    values that are not used for computation.
+
+    >>> in_values, out_values, rest_values = apply_to_iterator(
+    >>>     func, iterator, n_input)
+    >>> # in_values: (iter of in_val0, ..., iter of in_val{n_input - 1})
+    >>> # out_values: (iter of out_val0, ...)
+    >>> # rest_values: (iter of rest_val0, ...)
+
+    Here is an exmple, which applies a pretrained Faster R-CNN to
+    PASCAL VOC dataset.
+
+    >>> from chainer import iterators
+    >>>
+    >>> from chainercv.datasets import VOCBBoxDataset
+    >>> from chainercv.links import FasterRCNNVGG16
+    >>> from chainercv.utils import apply_to_iterator
+    >>>
+    >>> dataset = VOCBBoxDataset(year='2007', split='test')
+    >>> # next(iterator) -> [(img, gt_bbox, gt_label)]
+    >>> iterator = iterators.SerialIterator(
+    ...     dataset, 2, repeat=False, shuffle=False)
+    >>>
+    >>> # model.predict([img]) -> ([pred_bbox], [pred_label], [pred_score])
+    >>> model = FasterRCNNVGG16(pretrained_model='voc07')
+    >>>
+    >>> in_values, out_values, rest_values = apply_to_iterator(
+    ...     model.predict, iterator)
+    >>>
+    >>> # in_values contains one iterator
+    >>> imgs, = in_values
+    >>> # out_values contains three iterators
+    >>> pred_bboxes, pred_labels, pred_scores = out_values
+    >>> # rest_values contains two iterators
+    >>> gt_bboxes, gt_labels = rest_values
+
+    Args:
+        func: A callable that takes batch(es) of input data and returns
+            computed data.
+        iterator (iterator): An iterator of batches.
+            The first :obj:`n_input` elements in each sample are
+            treated as input values. They are passed to :obj:`func`.
+        n_input (int): The number of input data. The default value is :obj:`1`.
+        hook: A callable that is called after each iteration.
+            :obj:`in_values`, :obj:`out_values`, and :obj:`rest_values`
+            are passed as arguments.
+            Note that these values do not contain data from the previous
+            iterations.
+
+    Returns:
+        Three tuples of iterators:
+        This function returns three tuples of iterators:
+        :obj:`in_values`, :obj:`out_values` and :obj:`rest_values`.
+
+        * :obj:`in_values`: A tuple of iterators. Each iterator \
+            returns a corresponding input value. \
+            For example, if :func:`func` takes \
+            :obj:`[in_val0], [in_val1]`, :obj:`next(in_values[0])` \
+            and :obj:`next(in_values[1])` will be \
+            :obj:`in_val0` and :obj:`in_val1`.
+        * :obj:`out_values`: A tuple of iterators. Each iterator \
+            returns a corresponding computed value. \
+            For example, if :func:`func` returns \
+            :obj:`([out_val0], [out_val1])`, :obj:`next(out_values[0])` \
+            and :obj:`next(out_values[1])` will be \
+            :obj:`out_val0` and :obj:`out_val1`.
+        * :obj:`rest_values`: A tuple of iterators. Each iterator \
+            returns a corresponding rest value. \
+            For example, if the :obj:`iterator` returns \
+            :obj:`[(in_val0, in_val1, rest_val0, rest_val1)]`, \
+            :obj:`next(rest_values[0])` \
+            and :obj:`next(rest_values[1])` will be \
+            :obj:`rest_val0` and :obj:`rest_val1`. \
+            If the input \
+            iterator does not give any rest values, this tuple \
+            will be empty.
+    """
+
+    in_values, out_values, rest_values = unzip(
+        _apply(func, iterator, n_input, hook))
+
+    # in_values: iter of ([in_val0], [in_val1], ...)
+    #     -> (iter of in_val0, iter of in_val1, ...)
+    in_values = tuple(map(_flatten, unzip(in_values)))
+
+    # out_values: iter of ([out_val0], [out_val1], ...)
+    #     -> (iter of out_val0, iter of out_val1, ...)
+    out_values = tuple(map(_flatten, unzip(out_values)))
+
+    # rest_values: iter of ([rest_val0], [rest_val1], ...)
+    #     -> (iter of rest_val0, iter of rest_val1, ...)
+    rest_values = tuple(map(_flatten, unzip(rest_values)))
+
+    return in_values, out_values, rest_values
+
+
+def _apply(func, iterator, n_input, hook):
+    for batch in iterator:
+        # batch: [(in_val0, in_val1, ... , rest_val0, rest_val1, ...)] or
+        #     [in_val]
+
+        in_values = []
+        rest_values = []
+        for sample in batch:
+            if isinstance(sample, tuple):
+                in_values.append(sample[0:n_input])
+                rest_values.append(sample[n_input:])
+            else:
+                in_values.append((sample,))
+                rest_values.append(())
+
+        # in_values: [(in_val0, in_val1, ...)]
+        #     ->  ([in_val0], [in_val1], ...)
+        in_values = tuple(list(v) for v in zip(*in_values))
+
+        # rest_values: [(rest_val0, rest_val1, ...)]
+        #     -> ([rest_val0], [rest_val1], ...)
+        rest_values = tuple(list(v) for v in zip(*rest_values))
+
+        # out_values: ([out_val0], [out_val1], ...) or [out_val]
+        out_values = func(*in_values)
+        if not isinstance(out_values, tuple):
+            # pred_values: [out_val] -> ([out_val],)
+            out_values = out_values,
+
+        if hook:
+            hook(in_values, out_values, rest_values)
+
+        yield in_values, out_values, rest_values
+
+
+def _flatten(iterator):
+    return (sample for batch in iterator for sample in batch)