summary | shortlog | log | commit | commitdiff | tree
raw | patch | inline | side by side (from parent 1: f59720c)
raw | patch | inline | side by side (from parent 1: f59720c)
author | Xusong Wang <xusongw@google.com> | |
Thu, 4 Oct 2018 22:03:13 +0000 (15:03 -0700) | ||
committer | Przemyslaw Szczepaniak <pszczepaniak@google.com> | |
Wed, 10 Oct 2018 17:00:48 +0000 (17:00 +0000) |
Divide BBOX_TRANSFORM op into 2 ops
- AXIS_ALIGNED_BBOX_TRANSFORM
- ROTATED_BBOX_TRANSFORM
Rotated bounding boxes use different tensor format than axis-aligned
bounding boxes, and it would be less confusing if they were represented
by a separate operator code.
Bug: 113562630
Test: NeuralNetworksTest_static
Test: VtsHalNeuralnetworksV1_xTargetTest with sample driver
Change-Id: Ie08f2e0d0da77f6750766a394969653478d054d5
Merged-In: Ie08f2e0d0da77f6750766a394969653478d054d5
(cherry picked from commit d2bae1c268f8ef8e313441450e226d68ab515d9a)
- AXIS_ALIGNED_BBOX_TRANSFORM
- ROTATED_BBOX_TRANSFORM
Rotated bounding boxes use different tensor format than axis-aligned
bounding boxes, and it would be less confusing if they were represented
by a separate operator code.
Bug: 113562630
Test: NeuralNetworksTest_static
Test: VtsHalNeuralnetworksV1_xTargetTest with sample driver
Change-Id: Ie08f2e0d0da77f6750766a394969653478d054d5
Merged-In: Ie08f2e0d0da77f6750766a394969653478d054d5
(cherry picked from commit d2bae1c268f8ef8e313441450e226d68ab515d9a)
neuralnetworks/1.2/types.hal | patch | blob | history |
index 43771316d8a877e739435701fd69584d4458baa9..bed1d5cad90b5da76ab0e4e8d4d7f2ae5f790e60 100644 (file)
ARGMAX = 38,
ARGMIN = 39,
PAD_V2 = 40,
- BBOX_TRANSFORM = 41,
+ AXIS_ALIGNED_BBOX_TRANSFORM = 41,
BIDIRECTIONAL_SEQUENCE_LSTM = 42,
BIDIRECTIONAL_SEQUENCE_RNN = 43,
BOX_WITH_NMS_LIMIT = 44,
TRANSPOSE_CONV_2D = 84,
UNIDIRECTIONAL_SEQUENCE_LSTM = 85,
UNIDIRECTIONAL_SEQUENCE_RNN = 86,
+ ROTATED_BBOX_TRANSFORM = 87,
};
/**