COMPMID-1451: Fixes for BoundingBoxTransform
- Fixing a bug for which we did not scale the boxes before transforming them
- Adding the correct_transform_coords option to BoundingBoxTransformInfo
Change-Id: I40281254bcf87e7c8583c119e99562414fe59822
diff --git a/arm_compute/core/Types.h b/arm_compute/core/Types.h
index 4eb8129..38094ee 100644
--- a/arm_compute/core/Types.h
+++ b/arm_compute/core/Types.h
@@ -1209,21 +1209,24 @@
};
/** Bounding Box Transform information class */
-class BoundingBoxTransformInfo
+class BoundingBoxTransformInfo final
{
public:
/** Constructor
*
- * @param[in] img_width Width of the original image
- * @param[in] img_height Height, of the original image
- * @param[in] scale Scale of the original image
- * @param[in] apply_scale (Optional)Re-apply scaling after transforming the boxes. Defaults to false
- * @param[in] weights (Optional)Weights [wx, wy, ww, wh] for the deltas. Defaults to all ones
- * @param[in] bbox_xform_clip (Optional)Minimum bounding box width and height after bounding box transformation in log-space. Defaults to log(1000/16)
+ * @param[in] img_width Width of the original image
+ * @param[in] img_height Height, of the original image
+ * @param[in] scale Scale of the original image
+ * @param[in] apply_scale (Optional)Re-apply scaling after transforming the boxes. Defaults to false
+ * @param[in] weights (Optional)Weights [wx, wy, ww, wh] for the deltas. Defaults to all ones
+ * @param[in] correct_transform_coords (Optional)Correct bounding box transform coordinates. Defaults to false
+ * @param[in] bbox_xform_clip (Optional)Minimum bounding box width and height after bounding box transformation in log-space. Defaults to log(1000/16)
*/
- BoundingBoxTransformInfo(float img_width, float img_height, float scale, bool apply_scale = false, const std::array<float, 4> weights = { { 1.f, 1.f, 1.f, 1.f } }, float bbox_xform_clip =
- 4.135166556742356f)
- : _img_width(img_width), _img_height(img_height), _scale(scale), _apply_scale(apply_scale), _weights(weights), _bbox_xform_clip(bbox_xform_clip)
+ BoundingBoxTransformInfo(float img_width, float img_height, float scale, bool apply_scale = false, const std::array<float, 4> weights = { { 1.f, 1.f, 1.f, 1.f } }, bool correct_transform_coords =
+ false,
+ float bbox_xform_clip =
+ 4.135166556742356f)
+ : _img_width(img_width), _img_height(img_height), _scale(scale), _apply_scale(apply_scale), _correct_transform_coords(correct_transform_coords), _weights(weights), _bbox_xform_clip(bbox_xform_clip)
{
}
@@ -1257,11 +1260,17 @@
return _apply_scale;
}
+ bool correct_transform_coords() const
+ {
+ return _correct_transform_coords;
+ }
+
private:
float _img_width;
float _img_height;
float _scale;
bool _apply_scale;
+ bool _correct_transform_coords;
std::array<float, 4> _weights;
float _bbox_xform_clip;
};
diff --git a/src/core/CL/cl_kernels/bounding_box_transform.cl b/src/core/CL/cl_kernels/bounding_box_transform.cl
index 14a0fad..0972355 100644
--- a/src/core/CL/cl_kernels/bounding_box_transform.cl
+++ b/src/core/CL/cl_kernels/bounding_box_transform.cl
@@ -23,7 +23,7 @@
*/
#include "helpers.h"
-#if defined(DATA_TYPE) && defined(WEIGHT_X) && defined(WEIGHT_Y) && defined(WEIGHT_W) && defined(WEIGHT_H) && defined(IMG_WIDTH) && defined(IMG_HEIGHT) && defined(BOX_FIELDS) // Check for compile time constants
+#if defined(DATA_TYPE) && defined(WEIGHT_X) && defined(WEIGHT_Y) && defined(WEIGHT_W) && defined(WEIGHT_H) && defined(IMG_WIDTH) && defined(IMG_HEIGHT) && defined(BOX_FIELDS) && defined(SCALE_BEFORE) // Check for compile time constants
/** Perform a padded copy of input tensor to the output tensor. Padding values are defined at compile time
*
@@ -74,10 +74,12 @@
const DATA_TYPE halfone = (DATA_TYPE)0.5f;
const int py = get_global_id(1); // box
+ const VEC_DATA_TYPE(DATA_TYPE, 4)
+ scale_before = (VEC_DATA_TYPE(DATA_TYPE, 4))SCALE_BEFORE;
VEC_DATA_TYPE(DATA_TYPE, 4)
delta = vload4(0, (__global DATA_TYPE *)deltas.ptr);
const VEC_DATA_TYPE(DATA_TYPE, 4)
- box = vload4(0, (__global DATA_TYPE *)vector_offset(&boxes, BOX_FIELDS * py));
+ box = vload4(0, (__global DATA_TYPE *)vector_offset(&boxes, BOX_FIELDS * py)) / scale_before;
// Calculate width and centers of the old boxes
const VEC_DATA_TYPE(DATA_TYPE, 2)
@@ -106,13 +108,16 @@
// Calculate the coordinates of the new boxes
VEC_DATA_TYPE(DATA_TYPE, 4)
pred_box = pred_ctr.s0101 + sign * halfone * pred_dims.s0101;
+#ifdef OFFSET // Possibly adjust the predicted boxes
+ pred_box.s23 -= one;
+#endif // Possibly adjust the predicted boxes
pred_box = CLAMP(pred_box, min_values, max_values);
-#ifdef SCALE // Possibly scale the predicted boxes
- pred_box *= (VEC_DATA_TYPE(DATA_TYPE, 4))SCALE;
+#ifdef SCALE_AFTER // Possibly scale the predicted boxes
+ pred_box *= (VEC_DATA_TYPE(DATA_TYPE, 4))SCALE_AFTER;
#endif // Possibly scale the predicted boxes
// Store them into the output
vstore4(pred_box, 0, (__global DATA_TYPE *)pred_boxes.ptr);
}
-#endif // defined(DATA_TYPE) && defined(WEIGHT_X) && defined(WEIGHT_Y) && defined(WEIGHT_W) && defined(WEIGHT_H) && defined(IMG_WIDTH) && defined(IMG_HEIGHT) && defined(BOX_FIELDS)
+#endif // defined(DATA_TYPE) && defined(WEIGHT_X) && defined(WEIGHT_Y) && defined(WEIGHT_W) && defined(WEIGHT_H) && defined(IMG_WIDTH) && defined(IMG_HEIGHT) && defined(BOX_FIELDS) && defined(SCALE_BEFORE)
diff --git a/src/core/CL/kernels/CLBoundingBoxTransformKernel.cpp b/src/core/CL/kernels/CLBoundingBoxTransformKernel.cpp
index 09f3d33..bff28e3 100644
--- a/src/core/CL/kernels/CLBoundingBoxTransformKernel.cpp
+++ b/src/core/CL/kernels/CLBoundingBoxTransformKernel.cpp
@@ -39,7 +39,7 @@
{
namespace
{
-Status validate_arguments(const ITensorInfo *boxes, const ITensorInfo *pred_boxes, const ITensorInfo *deltas)
+Status validate_arguments(const ITensorInfo *boxes, const ITensorInfo *pred_boxes, const ITensorInfo *deltas, const BoundingBoxTransformInfo &info)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(boxes, pred_boxes, deltas);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(boxes, DataType::F32, DataType::F16);
@@ -56,6 +56,7 @@
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(pred_boxes, deltas);
ARM_COMPUTE_RETURN_ERROR_ON(pred_boxes->num_dimensions() > 2);
}
+ ARM_COMPUTE_RETURN_ERROR_ON(info.scale() <= 0);
return Status{};
}
} // namespace
@@ -70,6 +71,8 @@
ARM_COMPUTE_ERROR_ON_NULLPTR(boxes, pred_boxes, deltas);
auto_init_if_empty(*pred_boxes->info(), *deltas->info());
+ ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(boxes->info(), pred_boxes->info(), deltas->info(), info));
+
// Set instance variables
_boxes = boxes;
_pred_boxes = pred_boxes;
@@ -90,7 +93,9 @@
build_opts.add_option("-DIMG_WIDTH=" + support::cpp11::to_string(img_w));
build_opts.add_option("-DIMG_HEIGHT=" + support::cpp11::to_string(img_h));
build_opts.add_option("-DBOX_FIELDS=" + support::cpp11::to_string(4));
- build_opts.add_option_if(info.apply_scale(), "-DSCALE=" + float_to_string_with_full_precision(info.scale()));
+ build_opts.add_option("-DSCALE_BEFORE=" + float_to_string_with_full_precision(info.scale()));
+ build_opts.add_option_if(info.apply_scale(), "-DSCALE_AFTER=" + float_to_string_with_full_precision(info.scale()));
+ build_opts.add_option_if(info.correct_transform_coords(), "-DOFFSET=1");
// Create kernel
_kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("bounding_box_transform", build_opts.options()));
@@ -103,8 +108,7 @@
Status CLBoundingBoxTransformKernel::validate(const ITensorInfo *boxes, const ITensorInfo *pred_boxes, const ITensorInfo *deltas, const BoundingBoxTransformInfo &info)
{
- ARM_COMPUTE_UNUSED(info);
- ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(boxes, pred_boxes, deltas));
+ ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(boxes, pred_boxes, deltas, info));
return Status{};
}
diff --git a/tests/validation/CL/BoundingBoxTransform.cpp b/tests/validation/CL/BoundingBoxTransform.cpp
index dba3a0e..c5856ca 100644
--- a/tests/validation/CL/BoundingBoxTransform.cpp
+++ b/tests/validation/CL/BoundingBoxTransform.cpp
@@ -50,7 +50,8 @@
BoundingBoxTransformInfo(128U, 128U, 4U, true),
BoundingBoxTransformInfo(800U, 600U, 1U, false),
BoundingBoxTransformInfo(800U, 600U, 2U, true, { 1.0, 0.5, 1.5, 2.0 }),
- BoundingBoxTransformInfo(800U, 600U, 4U, false, { 1.0, 0.5, 1.5, 2.0 })
+ BoundingBoxTransformInfo(800U, 600U, 4U, false, { 1.0, 0.5, 1.5, 2.0 }),
+ BoundingBoxTransformInfo(800U, 600U, 4U, false, { 1.0, 0.5, 1.5, 2.0 }, true)
});
const auto DeltaDataset = framework::dataset::make("DeltasShape", { TensorShape(36U, 1U),
@@ -74,22 +75,26 @@
TensorInfo(TensorShape(5U, 128U), 1, DataType::F32), // Wrong number of box fields
TensorInfo(TensorShape(4U, 128U), 1, DataType::F16), // Wrong data type
TensorInfo(TensorShape(4U, 128U), 1, DataType::F32), // Wrong number of classes
- TensorInfo(TensorShape(4U, 128U), 1, DataType::F32)}), // Deltas and predicted boxes have different dimensions
+ TensorInfo(TensorShape(4U, 128U), 1, DataType::F32), // Deltas and predicted boxes have different dimensions
+ TensorInfo(TensorShape(4U, 128U), 1, DataType::F32)}), // Scaling is zero
framework::dataset::make("PredBoxesInfo",{ TensorInfo(TensorShape(128U, 128U), 1, DataType::F32),
TensorInfo(TensorShape(128U, 128U), 1, DataType::F32),
TensorInfo(TensorShape(127U, 128U), 1, DataType::F32),
TensorInfo(TensorShape(128U, 100U), 1, DataType::F32),
- TensorInfo(TensorShape(128U, 100U), 1, DataType::F32)})),
+ TensorInfo(TensorShape(128U, 100U), 1, DataType::F32),
+ TensorInfo(TensorShape(128U, 128U), 1, DataType::F32)})),
framework::dataset::make("DeltasInfo", { TensorInfo(TensorShape(128U, 128U), 1, DataType::F32),
TensorInfo(TensorShape(128U, 128U), 1, DataType::F32),
TensorInfo(TensorShape(127U, 128U), 1, DataType::F32),
TensorInfo(TensorShape(128U, 100U), 1, DataType::F32),
+ TensorInfo(TensorShape(128U, 128U), 1, DataType::F32),
TensorInfo(TensorShape(128U, 128U), 1, DataType::F32)})),
framework::dataset::make("BoundingBoxTransofmInfo", { BoundingBoxTransformInfo(800.f, 600.f, 1.f),
BoundingBoxTransformInfo(800.f, 600.f, 1.f),
BoundingBoxTransformInfo(800.f, 600.f, 1.f),
- BoundingBoxTransformInfo(800.f, 600.f, 1.f)})),
- framework::dataset::make("Expected", { true, false, false, false, false })),
+ BoundingBoxTransformInfo(800.f, 600.f, 1.f),
+ BoundingBoxTransformInfo(800.f, 600.f, 0.f)})),
+ framework::dataset::make("Expected", { true, false, false, false, false, false})),
boxes_info, pred_boxes_info, deltas_info, bbox_info, expected)
{
ARM_COMPUTE_EXPECT(bool(CLBoundingBoxTransform::validate(&boxes_info.clone()->set_is_resizable(true), &pred_boxes_info.clone()->set_is_resizable(true), &deltas_info.clone()->set_is_resizable(true), bbox_info)) == expected, framework::LogLevel::ERRORS);
diff --git a/tests/validation/reference/BoundingBoxTransform.cpp b/tests/validation/reference/BoundingBoxTransform.cpp
index 9918ff6..55dd165 100644
--- a/tests/validation/reference/BoundingBoxTransform.cpp
+++ b/tests/validation/reference/BoundingBoxTransform.cpp
@@ -50,7 +50,10 @@
const int img_h = floor(info.img_height() / info.scale() + 0.5f);
const int img_w = floor(info.img_width() / info.scale() + 0.5f);
- const T scale = (info.apply_scale() ? T(info.scale()) : T(1));
+ const auto scale_after = (info.apply_scale() ? T(info.scale()) : T(1));
+ const auto scale_before = T(info.scale());
+ ARM_COMPUTE_ERROR_ON(scale_before <= 0);
+ const auto offset = (info.correct_transform_coords() ? T(1.f) : T(0.f));
const size_t box_fields = 4;
const size_t class_fields = 4;
@@ -59,10 +62,10 @@
{
// Extract ROI information
const size_t start_box = box_fields * i;
- const T width = boxes[start_box + 2] - boxes[start_box] + T(1.0);
- const T height = boxes[start_box + 3] - boxes[start_box + 1] + T(1.0);
- const T ctr_x = boxes[start_box] + T(0.5) * width;
- const T ctr_y = boxes[start_box + 1] + T(0.5) * height;
+ const T width = (boxes[start_box + 2] / scale_before) - (boxes[start_box] / scale_before) + T(1.f);
+ const T height = (boxes[start_box + 3] / scale_before) - (boxes[start_box + 1] / scale_before) + T(1.f);
+ const T ctr_x = (boxes[start_box] / scale_before) + T(0.5f) * width;
+ const T ctr_y = (boxes[start_box + 1] / scale_before) + T(0.5f) * height;
for(size_t j = 0; j < num_classes; ++j)
{
@@ -84,10 +87,10 @@
const T pred_h = T(std::exp(dh)) * height;
// Store the prediction into the output tensor
- pred_boxes_ptr[start_delta] = scale * utility::clamp<T>(pred_ctr_x - T(0.5) * pred_w, T(0), T(img_w - 1));
- pred_boxes_ptr[start_delta + 1] = scale * utility::clamp<T>(pred_ctr_y - T(0.5) * pred_h, T(0), T(img_h - 1));
- pred_boxes_ptr[start_delta + 2] = scale * utility::clamp<T>(pred_ctr_x + T(0.5) * pred_w, T(0), T(img_w - 1));
- pred_boxes_ptr[start_delta + 3] = scale * utility::clamp<T>(pred_ctr_y + T(0.5) * pred_h, T(0), T(img_h - 1));
+ pred_boxes_ptr[start_delta] = scale_after * utility::clamp<T>(pred_ctr_x - T(0.5f) * pred_w, T(0), T(img_w - 1));
+ pred_boxes_ptr[start_delta + 1] = scale_after * utility::clamp<T>(pred_ctr_y - T(0.5f) * pred_h, T(0), T(img_h - 1));
+ pred_boxes_ptr[start_delta + 2] = scale_after * utility::clamp<T>(pred_ctr_x + T(0.5f) * pred_w - offset, T(0), T(img_w - 1));
+ pred_boxes_ptr[start_delta + 3] = scale_after * utility::clamp<T>(pred_ctr_y + T(0.5f) * pred_h - offset, T(0), T(img_h - 1));
}
}
return pred_boxes;