add rcnn loss tests

toshihiroryuu · Aug 25, 2017 · 2a1e299 · 2a1e299
1 parent 0d9c1db
commit 2a1e299
Showing 1 changed file with 53 additions and 2 deletions.
diff --git a/tests/layers/test_losses.py b/tests/layers/test_losses.py
@@ -3,7 +3,7 @@
 import numpy
 import keras_rcnn.layers.object_detection
 
-def test_call_classification():
+def test_rpn_classification():
     anchors = 9
     layer = keras_rcnn.layers.RPNClassificationLoss(anchors=anchors)
     y_true = keras.backend.variable(100 * numpy.random.random((1, 91, 4)))
@@ -19,7 +19,7 @@ def test_call_classification():
     assert numpy.isclose(keras.backend.eval(layer.losses[0]), expected_loss)
 
 
-def test_call_regression():
+def test_rpn_regression():
     anchors = 9
     metadata = keras.backend.variable(numpy.array([[224, 224, 1]]))
     layer = keras_rcnn.layers.RPNRegressionLoss(anchors=anchors)
@@ -41,3 +41,54 @@ def test_call_regression():
     assert len(layer.losses) == 1
 
     assert numpy.isclose(keras.backend.eval(layer.losses[0]), expected_loss)
+
+
+def test_rcnn_classification():
+    num_classes = 5
+    layer = keras_rcnn.layers.losses.RCNNClassificationLoss()
+    classes = numpy.random.choice(range(0, num_classes), (91))
+    target = numpy.zeros((1, 91, num_classes))
+    target[0, numpy.arange(91), classes] = 1
+    target = keras.backend.variable(target)
+    scores = keras.backend.variable(numpy.random.random((1, 91, num_classes)))
+
+    numpy.testing.assert_array_equal(layer.call([target, scores]), scores)
+
+    assert len(layer.losses) == 1
+
+
+def test_rcnn_regression():
+    num_classes = 5
+    layer = keras_rcnn.layers.losses.RCNNRegressionLoss()
+
+    deltas = numpy.zeros((1, 91, 4 * num_classes))
+    target = numpy.zeros((1, 91, 4 * num_classes))
+
+    expected_loss = 0
+
+    numpy.testing.assert_array_equal(layer.call([target, deltas]), deltas)
+
+    assert len(layer.losses) == 1
+
+    assert numpy.isclose(keras.backend.eval(layer.losses[0]), expected_loss)
+
+    deltas = numpy.array([[1, 0, 1, 0, 3, 4, 5, 6, 0, 0, 0, 0],
+                          [0, 0, 0, 0, 7, 3, 5, 6, 4, 1, 1, 5],
+                          [5, 2, 0, 0, 2, 0, 7, 8, 1, 5, 2, 10],
+                          [0, 0, 0, 0, 5, 3, 13, 4, 0, 0, 0, 0],
+                          [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]])
+    deltas = numpy.expand_dims(deltas, 0)
+    deltas = keras.backend.variable(deltas)
+    target = numpy.array([[0, 0, 0, 0, 3, 4, 5, 6, 0, 0, 0, 0],
+                          [0, 0, 0, 0, 7, 3, 5, 6, 0, 0, 0, 0],
+                          [0, 0, 0, 0, 0, 0, 0, 0, 1, 5, 2, 10],
+                          [0, 0, 0, 0, 5, 3, 13, 4, 0, 0, 0, 0],
+                          [0, 0, 0, 0, 0, 0, 0, 0, 40, 12, 2, .5]])
+    target = numpy.expand_dims(target, 0)
+    target = keras.backend.variable(target)
+
+    numpy.testing.assert_array_equal(layer.call([target, deltas]), deltas)
+
+    expected_loss = 52.625
+
+    assert keras.backend.eval(layer.losses[1]) == expected_loss