rename psroi_pooling_2d -> ps_roi_average_pooling_2d

chainercv/experimental/links/model/fcis/fcis_resnet101.py

@@ -6,7 +6,7 @@
 import numpy as np
 
 from chainercv.experimental.links.model.fcis import FCIS
-from chainercv.functions import psroi_pooling_2d
+from chainercv.functions import ps_roi_average_pooling_2d
 from chainercv.links import Conv2DBNActiv
 from chainercv.links.model.faster_rcnn.region_proposal_network import \
     RegionProposalNetwork
@@ -365,7 +365,7 @@ def _pool(
             self, h_cls_seg, h_ag_loc, rois, roi_indices, gt_roi_labels):
         # PSROI Pooling
         # shape: (n_roi, n_class, 2, roi_size, roi_size)
-        roi_cls_ag_seg_scores = psroi_pooling_2d(
+        roi_cls_ag_seg_scores = ps_roi_average_pooling_2d(
             h_cls_seg, rois, roi_indices,
             self.n_class * 2, self.roi_size, self.roi_size,
             self.spatial_scale, self.group_size)
@@ -374,7 +374,7 @@ def _pool(
             (-1, self.n_class, 2, self.roi_size, self.roi_size))
 
         # shape: (n_roi, 2*4, roi_size, roi_size)
-        roi_ag_loc_scores = psroi_pooling_2d(
+        roi_ag_loc_scores = ps_roi_average_pooling_2d(
             h_ag_loc, rois, roi_indices,
             2 * 4, self.roi_size, self.roi_size,
             self.spatial_scale, self.group_size)

chainercv/functions/__init__.py

@@ -1,2 +1 @@
-from chainercv.functions.psroi_pooling_2d import psroi_pooling_2d  # NOQA
-from chainercv.functions.psroi_pooling_2d import PSROIPooling2D  # NOQA
+from chainercv.functions.ps_roi_average_pooling_2d import ps_roi_average_pooling_2d  # NOQA

chainercv/functions/psroi_pooling_2d.py → chainercv/functions/ps_roi_average_pooling_2d.py

@@ -47,7 +47,7 @@ def _roi_pooling_slice(size, stride, max_size, roi_offset):
     return slice(start, end), end - start
 
 
-class PSROIPooling2D(function.Function):
+class PSROIAveragePooling2D(function.Function):
 
     def __init__(self, out_c, out_h, out_w, spatial_scale, group_size):
         self.out_c, self.out_h, self.out_w = out_c, out_h, out_w
@@ -197,7 +197,7 @@ def forward_gpu(self, inputs):
 
             float bin_area = (hend - hstart) * (wend - wstart);
             top_data = is_empty? (float) 0. : out_sum / bin_area;
-            ''', 'psroi_pooling_2d_fwd'
+            ''', 'ps_roi_average_pooling_2d_fwd'
         )(bottom_data, bottom_rois, bottom_roi_indices,
           self.spatial_scale, channels, height, width,
           self.out_c, self.out_h, self.out_w, self.group_size,
@@ -327,7 +327,7 @@ def backward_gpu(self, inputs, gy):
                     &bottom_diff[bottom_diff_offset + bottom_index], diff_val);
               }
             }
-            ''', 'psroi_pooling_2d_bwd'
+            ''', 'ps_roi_average_pooling_2d_bwd'
         )(gy[0], bottom_rois, bottom_roi_indices,
           self.spatial_scale, channels, height, width,
           self.out_c, self.out_h, self.out_w,
@@ -336,11 +336,11 @@ def backward_gpu(self, inputs, gy):
         return bottom_diff, None, None
 
 
-def psroi_pooling_2d(
+def ps_roi_average_pooling_2d(
         x, rois, roi_indices, out_c, out_h, out_w,
         spatial_scale, group_size
 ):
-    """Position Sensitive Region of Interest (ROI) pooling function.
+    """Position Sensitive Region of Interest (ROI) Average pooling function.
 
     This function computes position sensitive average of input spatial patch
     with the given region of interests. Each ROI is splitted into
@@ -368,5 +368,5 @@ def psroi_pooling_2d(
     `R-FCN <https://arxiv.org/abs/1605.06409>`_.
 
     """
-    return PSROIPooling2D(out_c, out_h, out_w, spatial_scale,
-                          group_size)(x, rois, roi_indices)
+    return PSROIAveragePooling2D(out_c, out_h, out_w, spatial_scale,
+                                 group_size)(x, rois, roi_indices)

tests/functions_tests/test_psroi_pooling_2d.py → tests/functions_tests/test_ps_roi_average_pooling_2d.py

@@ -10,7 +10,7 @@
 from chainercv import functions
 
 
-class TestPSROIPolling2D(unittest.TestCase):
+class TestPSROIAveragePolling2D(unittest.TestCase):
 
     def setUp(self):
         self.N = 3
@@ -43,7 +43,7 @@ def check_forward(self, x_data, roi_data, roi_index_data):
         x = chainer.Variable(x_data)
         rois = chainer.Variable(roi_data)
         roi_indices = chainer.Variable(roi_index_data)
-        y = functions.psroi_pooling_2d(
+        y = functions.ps_roi_average_pooling_2d(
             x, rois, roi_indices, self.out_c, self.out_h, self.out_w,
             self.spatial_scale, self.group_size)
         self.assertEqual(y.data.dtype, np.float32)
@@ -63,11 +63,12 @@ def test_forward_gpu(self):
             cuda.to_gpu(self.roi_indices))
 
     def check_backward(self, x_data, roi_data, roi_index_data, y_grad_data):
+        def f(x, rois, roi_indices):
+            return functions.ps_roi_average_pooling_2d(
+                x, rois, roi_indices, self.out_c, self.out_h, self.out_w,
+                self.spatial_scale, self.group_size)
         gradient_check.check_backward(
-            functions.PSROIPooling2D(
-                self.out_c, self.out_h, self.out_w,
-                self.spatial_scale, self.group_size),
-            (x_data, roi_data, roi_index_data), y_grad_data,
+            f, (x_data, roi_data, roi_index_data), y_grad_data,
             no_grads=[False, True, True], **self.check_backward_options)
 
     @condition.retry(3)
@@ -85,7 +86,7 @@ def apply_backward(self, x_data, roi_data, roi_index_data, y_grad_data):
         x = chainer.Variable(x_data)
         rois = chainer.Variable(roi_data)
         roi_indices = chainer.Variable(roi_index_data)
-        y = functions.psroi_pooling_2d(
+        y = functions.ps_roi_average_pooling_2d(
             x, rois, roi_indices, self.out_c, self.out_h, self.out_w,
             self.spatial_scale, self.group_size)
         x.cleargrad()