Fixes and tests for hourglass variants.

PiperOrigin-RevId: 331166835

research/object_detection/models/keras_models/hourglass_network.py

@@ -226,7 +226,12 @@ def _make_repeated_residual_blocks(out_channels, num_blocks,
     residual_channels = out_channels
 
   for i in range(num_blocks - 1):
+    # Only use the stride at the first block so we don't repeatedly downsample
+    # the input
     stride = initial_stride if i == 0 else 1
+
+    # If the stide is more than 1, we cannot use an identity layer for the
+    # skip connection and are forced to use a conv for the skip connection.
     skip_conv = stride > 1
 
     blocks.append(
@@ -234,8 +239,18 @@ def _make_repeated_residual_blocks(out_channels, num_blocks,
                       skip_conv=skip_conv)
     )
 
+  if num_blocks == 1:
+    # If there is only 1 block, the for loop above is not run,
+    # therefore we honor the requested stride in the last residual block
+    stride = initial_stride
+    # We are forced to use a conv in the skip connection if stride > 1
+    skip_conv = stride > 1
+  else:
+    stride = 1
     skip_conv = residual_channels != out_channels
-  blocks.append(ResidualBlock(out_channels=out_channels, skip_conv=skip_conv))
+
+  blocks.append(ResidualBlock(out_channels=out_channels, skip_conv=skip_conv,
+                              stride=stride))
 
   return blocks
 
@@ -494,7 +509,7 @@ def hourglass_104():
   )
 
 
-def single_stage_hourglass(blocks_per_stage, num_channels):
+def single_stage_hourglass(blocks_per_stage, num_channels, downsample=True):
   nc = num_channels
   channel_dims = [nc, nc * 2, nc * 2, nc * 3, nc * 3, nc * 3, nc  * 4]
   num_stages = len(blocks_per_stage) - 1
@@ -504,20 +519,21 @@ def single_stage_hourglass(blocks_per_stage, num_channels):
       num_hourglasses=1,
       num_stages=num_stages,
       blocks_per_stage=blocks_per_stage,
+      downsample=downsample
   )
 
 
-def hourglass_10(num_channels):
-  return single_stage_hourglass([1, 1], num_channels)
+def hourglass_10(num_channels, downsample=True):
+  return single_stage_hourglass([1, 1], num_channels, downsample)
 
 
-def hourglass_20(num_channels):
-  return single_stage_hourglass([1, 1, 1, 2], num_channels)
+def hourglass_20(num_channels, downsample=True):
+  return single_stage_hourglass([1, 2, 2], num_channels, downsample)
 
 
-def hourglass_32(num_channels):
-  return single_stage_hourglass([1, 1, 2, 2, 2], num_channels)
+def hourglass_32(num_channels, downsample=True):
+  return single_stage_hourglass([2, 2, 2, 2], num_channels, downsample)
 
 
-def hourglass_52(num_channels):
-  return single_stage_hourglass([2, 2, 2, 2, 2, 4], num_channels)
+def hourglass_52(num_channels, downsample=True):
+  return single_stage_hourglass([2, 2, 2, 2, 2, 4], num_channels, downsample)

research/object_detection/models/keras_models/hourglass_network_tf2_test.py

@@ -111,21 +111,34 @@ class HourglassDepthTest(tf.test.TestCase):
     self.assertEqual(hourglass.hourglass_depth(net), 104)
 
   def test_hourglass_10(self):
-    net = hourglass.hourglass_10(2)
+    net = hourglass.hourglass_10(2, downsample=False)
     self.assertEqual(hourglass.hourglass_depth(net), 10)
 
+    outputs = net(tf.zeros((2, 32, 32, 3)))
+    self.assertEqual(outputs[0].shape, (2, 32, 32, 4))
+
   def test_hourglass_20(self):
-    net = hourglass.hourglass_20(2)
+    net = hourglass.hourglass_20(2, downsample=False)
     self.assertEqual(hourglass.hourglass_depth(net), 20)
 
+    outputs = net(tf.zeros((2, 32, 32, 3)))
+    self.assertEqual(outputs[0].shape, (2, 32, 32, 4))
+
   def test_hourglass_32(self):
-    net = hourglass.hourglass_32(2)
+    net = hourglass.hourglass_32(2, downsample=False)
     self.assertEqual(hourglass.hourglass_depth(net), 32)
 
+    outputs = net(tf.zeros((2, 32, 32, 3)))
+    self.assertEqual(outputs[0].shape, (2, 32, 32, 4))
+
   def test_hourglass_52(self):
-    net = hourglass.hourglass_52(2)
+    net = hourglass.hourglass_52(2, downsample=False)
     self.assertEqual(hourglass.hourglass_depth(net), 52)
 
+    outputs = net(tf.zeros((2, 32, 32, 3)))
+    self.assertEqual(outputs[0].shape, (2, 32, 32, 4))
+
 
 if __name__ == '__main__':
   tf.test.main()
+