[Feature]: Add Rotated ROI align op for pytorch (cpu&cuda), parrots (…

…cpu&cuda) and onnxruntime (cpu) (#933)

* add roi_align_rotated

* code format

* Add align key to roi align rotated

* Add clockwise for rotated roi align

* fix bugs in onnx export

* Add docstring for RoIAlignRotated

* remove cuda unittest

* Reformat c++ code

* add onnx roi align rotated file

* fix unittest

* Add  cpu and float64 of cuda support for parrots

* code format

* Add unified header to roi align rotated

Co-authored-by: luopeichao <luopeichao@sensetime.com>

mmcv/ops/__init__.py

@@ -24,6 +24,7 @@
                            rel_roi_point_to_rel_img_point)
 from .psa_mask import PSAMask
 from .roi_align import RoIAlign, roi_align
+from .roi_align_rotated import RoIAlignRotated, roi_align_rotated
 from .roi_pool import RoIPool, roi_pool
 from .saconv import SAConv2d
 from .sync_bn import SyncBatchNorm
@@ -44,5 +45,6 @@
     'ConvTranspose2d', 'Linear', 'MaxPool2d', 'CrissCrossAttention', 'PSAMask',
     'point_sample', 'rel_roi_point_to_rel_img_point', 'SimpleRoIAlign',
     'SAConv2d', 'TINShift', 'tin_shift', 'box_iou_rotated', 'nms_rotated',
-    'upfirdn2d', 'FusedBiasLeakyReLU', 'fused_bias_leakyrelu'
+    'upfirdn2d', 'FusedBiasLeakyReLU', 'fused_bias_leakyrelu',
+    'RoIAlignRotated', 'roi_align_rotated'
 ]

mmcv/ops/box_iou_rotated.py

@@ -16,8 +16,10 @@ def box_iou_rotated(bboxes1, bboxes2, mode='iou', aligned=False):
     Arguments:
         boxes1 (Tensor): rotated bboxes 1. \
             It has shape (N, 5), indicating (x, y, w, h, theta) for each row.
+            Note that theta is in radian.
         boxes2 (Tensor): rotated bboxes 2. \
             It has shape (M, 5), indicating (x, y, w, h, theta) for each row.
+            Note that theta is in radian.
         mode (str): "iou" (intersection over union) or iof (intersection over
             foreground).
 

mmcv/ops/csrc/onnxruntime/cpu/onnxruntime_register.cpp

@@ -4,12 +4,14 @@
 #include "nms.h"
 #include "ort_mmcv_utils.h"
 #include "roi_align.h"
+#include "roi_align_rotated.h"
 #include "soft_nms.h"
 
 const char *c_MMCVOpDomain = "mmcv";
 SoftNmsOp c_SoftNmsOp;
 NmsOp c_NmsOp;
 MMCVRoiAlignCustomOp c_MMCVRoiAlignCustomOp;
+MMCVRoIAlignRotatedCustomOp c_MMCVRoIAlignRotatedCustomOp;
 GridSampleOp c_GridSampleOp;
 
 OrtStatus *ORT_API_CALL RegisterCustomOps(OrtSessionOptions *options,
@@ -34,6 +36,11 @@ OrtStatus *ORT_API_CALL RegisterCustomOps(OrtSessionOptions *options,
     return status;
   }
 
+  if (auto status =
+          ortApi->CustomOpDomain_Add(domain, &c_MMCVRoIAlignRotatedCustomOp)) {
+    return status;
+  }
+
   if (auto status = ortApi->CustomOpDomain_Add(domain, &c_GridSampleOp)) {
     return status;
   }

mmcv/ops/csrc/onnxruntime/cpu/roi_align_rotated.cpp

@@ -0,0 +1,246 @@
+// Modified from
+// https://github.com/facebookresearch/detectron2/tree/master/detectron2/layers/csrc/ROIAlignRotated
+// Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
+#include "roi_align_rotated.h"
+#include "../ort_mmcv_utils.h"
+
+struct PreCalc {
+  int pos1;
+  int pos2;
+  int pos3;
+  int pos4;
+  float w1;
+  float w2;
+  float w3;
+  float w4;
+};
+
+void pre_calc_for_bilinear_interpolate(
+    const int height, const int width, const int pooled_height,
+    const int pooled_width, const int iy_upper, const int ix_upper,
+    float roi_start_h, float roi_start_w, float bin_size_h, float bin_size_w,
+    int roi_bin_grid_h, int roi_bin_grid_w, float roi_center_h,
+    float roi_center_w, float cos_theta, float sin_theta,
+    std::vector<PreCalc> &pre_calc) {
+  int pre_calc_index = 0;
+  for (int ph = 0; ph < pooled_height; ph++) {
+    for (int pw = 0; pw < pooled_width; pw++) {
+      for (int iy = 0; iy < iy_upper; iy++) {
+        const float yy =
+            roi_start_h + ph * bin_size_h +
+            static_cast<float>(iy + .5f) * bin_size_h /
+                static_cast<float>(roi_bin_grid_h);  // e.g., 0.5, 1.5
+        for (int ix = 0; ix < ix_upper; ix++) {
+          const float xx = roi_start_w + pw * bin_size_w +
+                           static_cast<float>(ix + .5f) * bin_size_w /
+                               static_cast<float>(roi_bin_grid_w);
+
+          // Rotate by theta around the center and translate
+          // In image space, (y, x) is the order for Right Handed System,
+          // and this is essentially multiplying the point by a rotation matrix
+          // to rotate it counterclockwise through angle theta.
+          float y = yy * cos_theta - xx * sin_theta + roi_center_h;
+          float x = yy * sin_theta + xx * cos_theta + roi_center_w;
+          // deal with: inverse elements are out of feature map boundary
+          if (y < -1.0 || y > height || x < -1.0 || x > width) {
+            // empty
+            PreCalc pc;
+            pc.pos1 = 0;
+            pc.pos2 = 0;
+            pc.pos3 = 0;
+            pc.pos4 = 0;
+            pc.w1 = 0;
+            pc.w2 = 0;
+            pc.w3 = 0;
+            pc.w4 = 0;
+            pre_calc[pre_calc_index] = pc;
+            pre_calc_index += 1;
+            continue;
+          }
+
+          if (y < 0) {
+            y = 0;
+          }
+          if (x < 0) {
+            x = 0;
+          }
+
+          int y_low = (int)y;
+          int x_low = (int)x;
+          int y_high;
+          int x_high;
+
+          if (y_low >= height - 1) {
+            y_high = y_low = height - 1;
+            y = (float)y_low;
+          } else {
+            y_high = y_low + 1;
+          }
+
+          if (x_low >= width - 1) {
+            x_high = x_low = width - 1;
+            x = (float)x_low;
+          } else {
+            x_high = x_low + 1;
+          }
+
+          float ly = y - y_low;
+          float lx = x - x_low;
+          float hy = 1. - ly, hx = 1. - lx;
+          float w1 = hy * hx, w2 = hy * lx, w3 = ly * hx, w4 = ly * lx;
+
+          // save weights and indices
+          PreCalc pc;
+          pc.pos1 = y_low * width + x_low;
+          pc.pos2 = y_low * width + x_high;
+          pc.pos3 = y_high * width + x_low;
+          pc.pos4 = y_high * width + x_high;
+          pc.w1 = w1;
+          pc.w2 = w2;
+          pc.w3 = w3;
+          pc.w4 = w4;
+          pre_calc[pre_calc_index] = pc;
+
+          pre_calc_index += 1;
+        }
+      }
+    }
+  }
+}
+
+void ROIAlignRotatedForwardCPU(const int nthreads, const float *input,
+                               const float *rois, float *output,
+                               const float &spatial_scale, const int aligned,
+                               const int clockwise, const int channels,
+                               const int height, const int width,
+                               const int pooled_height, const int pooled_width,
+                               const int sampling_ratio) {
+  int n_rois = nthreads / channels / pooled_width / pooled_height;
+  // (n, c, ph, pw) is an element in the pooled output
+  // can be parallelized using omp
+  // #pragma omp parallel for num_threads(32)
+  for (int n = 0; n < n_rois; n++) {
+    int index_n = n * channels * pooled_width * pooled_height;
+
+    const float *current_roi = rois + n * 6;
+    int roi_batch_ind = current_roi[0];
+
+    // Do not use rounding; this implementation detail is critical
+    float offset = aligned ? (float)0.5 : (float)0.0;
+    float roi_center_w = current_roi[1] * spatial_scale - offset;
+    float roi_center_h = current_roi[2] * spatial_scale - offset;
+    float roi_width = current_roi[3] * spatial_scale;
+    float roi_height = current_roi[4] * spatial_scale;
+    // float theta = current_roi[5] * M_PI / 180.0;
+    float theta = current_roi[5];  // Radian angle by default
+    if (clockwise) {
+      theta = -theta;
+    }
+    float cos_theta = cos(theta);
+    float sin_theta = sin(theta);
+    if (!aligned) {  // for backward-compatibility only
+      roi_width = std::max(roi_width, (float)1.);
+      roi_height = std::max(roi_height, (float)1.);
+    }
+
+    float bin_size_h =
+        static_cast<float>(roi_height) / static_cast<float>(pooled_height);
+    float bin_size_w =
+        static_cast<float>(roi_width) / static_cast<float>(pooled_width);
+
+    // We use roi_bin_grid to sample the grid and mimic integral
+    int roi_bin_grid_h = (sampling_ratio > 0)
+                             ? sampling_ratio
+                             : ceil(roi_height / pooled_height);  // e.g., = 2
+    int roi_bin_grid_w =
+        (sampling_ratio > 0) ? sampling_ratio : ceil(roi_width / pooled_width);
+
+    // We do average (integral) pooling inside a bin
+    const float count =
+        std::max(roi_bin_grid_h * roi_bin_grid_w, 1);  // e.g. = 4
+
+    // we want to precalculate indices and weights shared by all channels,
+    // this is the key point of optimization
+    std::vector<PreCalc> pre_calc(roi_bin_grid_h * roi_bin_grid_w *
+                                  pooled_width * pooled_height);
+
+    // roi_start_h and roi_start_w are computed wrt the center of RoI (x, y).
+    // Appropriate translation needs to be applied after.
+    float roi_start_h = -roi_height / 2.0;
+    float roi_start_w = -roi_width / 2.0;
+
+    pre_calc_for_bilinear_interpolate(
+        height, width, pooled_height, pooled_width, roi_bin_grid_h,
+        roi_bin_grid_w, roi_start_h, roi_start_w, bin_size_h, bin_size_w,
+        roi_bin_grid_h, roi_bin_grid_w, roi_center_h, roi_center_w, cos_theta,
+        sin_theta, pre_calc);
+
+    for (int c = 0; c < channels; c++) {
+      int index_n_c = index_n + c * pooled_width * pooled_height;
+      const float *offset_input =
+          input + (roi_batch_ind * channels + c) * height * width;
+      int pre_calc_index = 0;
+
+      for (int ph = 0; ph < pooled_height; ph++) {
+        for (int pw = 0; pw < pooled_width; pw++) {
+          int index = index_n_c + ph * pooled_width + pw;
+
+          float output_val = 0.;
+          for (int iy = 0; iy < roi_bin_grid_h; iy++) {
+            for (int ix = 0; ix < roi_bin_grid_w; ix++) {
+              PreCalc pc = pre_calc[pre_calc_index];
+              output_val += pc.w1 * offset_input[pc.pos1] +
+                            pc.w2 * offset_input[pc.pos2] +
+                            pc.w3 * offset_input[pc.pos3] +
+                            pc.w4 * offset_input[pc.pos4];
+
+              pre_calc_index += 1;
+            }
+          }
+          output_val /= count;
+
+          output[index] = output_val;
+        }  // for pw
+      }    // for ph
+    }      // for c
+  }        // for n
+}
+
+void MMCVRoIAlignRotatedKernel::Compute(OrtKernelContext *context) {
+  // Setup inputs
+  const OrtValue *input_X = ort_.KernelContext_GetInput(context, 0);
+  const float *X_data =
+      reinterpret_cast<const float *>(ort_.GetTensorData<float>(input_X));
+  const OrtValue *input_rois = ort_.KernelContext_GetInput(context, 1);
+  const float *rois = reinterpret_cast<const float *>(
+      ort_.GetTensorData<const float *>(input_rois));
+
+  // Setup output
+  OrtTensorDimensions out_dimensions(ort_, input_X);
+  OrtTensorDimensions roi_dimensions(ort_, input_rois);
+
+  int batch_size = out_dimensions.data()[0];
+  int input_channels = out_dimensions.data()[1];
+  int input_height = out_dimensions.data()[2];
+  int input_width = out_dimensions.data()[3];
+
+  out_dimensions.data()[0] = roi_dimensions.data()[0];
+  out_dimensions.data()[2] = aligned_height_;
+  out_dimensions.data()[3] = aligned_width_;
+
+  OrtValue *output = ort_.KernelContext_GetOutput(
+      context, 0, out_dimensions.data(), out_dimensions.size());
+  float *out = ort_.GetTensorMutableData<float>(output);
+  OrtTensorTypeAndShapeInfo *output_info = ort_.GetTensorTypeAndShape(output);
+  ort_.ReleaseTensorTypeAndShapeInfo(output_info);
+
+  // TODO: forward here
+  int output_size = out_dimensions.data()[0];
+  for (auto i = 1; i < out_dimensions.size(); ++i) {
+    output_size *= out_dimensions.data()[i];
+  }
+  ROIAlignRotatedForwardCPU(output_size, X_data, rois, out, spatial_scale_,
+                            aligned_, clockwise_, input_channels, input_height,
+                            input_width, aligned_height_, aligned_width_,
+                            sampling_ratio_);
+}

mmcv/ops/csrc/onnxruntime/roi_align_rotated.h

@@ -0,0 +1,61 @@
+#ifndef ONNXRUNTIME_ROI_ALIGN_ROTATED_H
+#define ONNXRUNTIME_ROI_ALIGN_ROTATED_H
+
+#include <assert.h>
+#include <onnxruntime_cxx_api.h>
+
+#include <cmath>
+#include <mutex>
+#include <string>
+#include <vector>
+
+struct MMCVRoIAlignRotatedKernel {
+ public:
+  MMCVRoIAlignRotatedKernel(Ort::CustomOpApi ort, const OrtKernelInfo* info)
+      : ort_(ort) {
+    aligned_height_ =
+        ort_.KernelInfoGetAttribute<int64_t>(info, "output_height");
+    aligned_width_ = ort_.KernelInfoGetAttribute<int64_t>(info, "output_width");
+    sampling_ratio_ =
+        ort_.KernelInfoGetAttribute<int64_t>(info, "sampling_ratio");
+    spatial_scale_ = ort_.KernelInfoGetAttribute<float>(info, "spatial_scale");
+    aligned_ = ort_.KernelInfoGetAttribute<int64_t>(info, "aligned");
+    clockwise_ = ort_.KernelInfoGetAttribute<int64_t>(info, "clockwise");
+  }
+
+  void Compute(OrtKernelContext* context);
+
+ private:
+  Ort::CustomOpApi ort_;
+  int aligned_height_;
+  int aligned_width_;
+  float spatial_scale_;
+  int sampling_ratio_;
+  int aligned_;
+  int clockwise_;
+};
+
+struct MMCVRoIAlignRotatedCustomOp
+    : Ort::CustomOpBase<MMCVRoIAlignRotatedCustomOp,
+                        MMCVRoIAlignRotatedKernel> {
+  void* CreateKernel(Ort::CustomOpApi api, const OrtKernelInfo* info) {
+    return new MMCVRoIAlignRotatedKernel(api, info);
+  }
+  const char* GetName() const { return "MMCVRoIAlignRotated"; }
+
+  size_t GetInputTypeCount() const { return 2; }
+  ONNXTensorElementDataType GetInputType(size_t) const {
+    return ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT;
+  }
+
+  size_t GetOutputTypeCount() const { return 1; }
+  ONNXTensorElementDataType GetOutputType(size_t) const {
+    return ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT;
+  }
+
+  // force cpu
+  const char* GetExecutionProviderType() const {
+    return "CPUExecutionProvider";
+  }
+};
+#endif  // ONNXRUNTIME_ROI_ALIGN_ROTATED_H