soft_dvpp_decode_resize_jpeg_op issue resolved, input size can be an int,...

soft_dvpp_decode_resize_jpeg_op issue resolved, input size can be an int, break the constraints of input size being sequence (height, width)

mindspore/ccsrc/minddata/dataset/kernels/image/soft_dvpp/soft_dvpp_decode_random_crop_resize_jpeg_op.cc

@@ -43,8 +43,8 @@ Status SoftDvppDecodeRandomCropResizeJpegOp::GetCropInfo(const std::shared_ptr<T
   RETURN_IF_NOT_OK(GetCropBox(img_height, img_width, &x, &y, &crop_heigh, &crop_widht));
   crop_info->left = x;
   crop_info->up = y;
-  crop_info->right = crop_info->left + crop_widht;
-  crop_info->down = crop_info->up + crop_heigh;
+  crop_info->right = crop_info->left + crop_widht - 1;
+  crop_info->down = crop_info->up + crop_heigh - 1;
   return Status::OK();
 }
 

mindspore/ccsrc/minddata/dataset/kernels/image/soft_dvpp/soft_dvpp_decode_resize_jpeg_op.cc

@@ -15,6 +15,7 @@
  */
 #include "minddata/dataset/kernels/image/soft_dvpp/soft_dvpp_decode_resize_jpeg_op.h"
 #include <string>
+#include <vector>
 
 #include "./utils/external_soft_dp.h"
 #include "opencv2/opencv.hpp"
@@ -35,14 +36,32 @@ Status SoftDvppDecodeResizeJpegOp::Compute(const std::shared_ptr<Tensor> &input,
     SoftDpProcsessInfo info;
     info.input_buffer = static_cast<uint8_t *>(buffer);
     info.input_buffer_size = input->SizeInBytes();
-    info.output_width = target_width_;
-    info.output_height = target_height_;
+
+    int input_w = 0;
+    int input_h = 0;
+    RETURN_IF_NOT_OK(GetJpegImageInfo(input, &input_w, &input_h));
 
     SoftDpCropInfo crop_info{0, 0, 0, 0};
 
-    cv::Mat out_rgb_img(target_height_, target_width_, CV_8UC3);
+    if (target_width_ == 0) {
+      if (input_h < input_w) {
+        CHECK_FAIL_RETURN_UNEXPECTED(input_h != 0, "The input height is 0");
+        info.output_height = target_height_;
+        info.output_width = static_cast<int>(std::lround(static_cast<float>(input_w) / input_h * info.output_height));
+      } else {
+        CHECK_FAIL_RETURN_UNEXPECTED(input_w != 0, "The input width is 0");
+        info.output_width = target_height_;
+        info.output_height = static_cast<int>(std::lround(static_cast<float>(input_h) / input_w * info.output_width));
+      }
+    } else {
+      info.output_height = target_height_;
+      info.output_width = target_width_;
+    }
+
+    cv::Mat out_rgb_img(info.output_height, info.output_width, CV_8UC3);
     info.output_buffer = out_rgb_img.data;
-    info.output_buffer_size = target_width_ * target_height_ * 3;
+    info.output_buffer_size = info.output_height * info.output_width * 3;
+
     info.is_v_before_u = true;
     int ret = DecodeAndResizeJpeg(&info);
     std::string error_info("Soft dvpp DecodeAndResizeJpeg failed with return code: ");
@@ -56,5 +75,16 @@ Status SoftDvppDecodeResizeJpegOp::Compute(const std::shared_ptr<Tensor> &input,
   }
   return Status::OK();
 }
+
+Status SoftDvppDecodeResizeJpegOp::OutputShape(const std::vector<TensorShape> &inputs,
+                                               std::vector<TensorShape> &outputs) {
+  RETURN_IF_NOT_OK(TensorOp::OutputShape(inputs, outputs));
+  outputs.clear();
+  TensorShape out({-1, -1, 3});  // we don't know what is output image size, but we know it should be 3 channels
+  if (inputs[0].Rank() == 1) outputs.emplace_back(out);
+  if (!outputs.empty()) return Status::OK();
+  return Status(StatusCode::kUnexpectedError, "Input has a wrong shape");
+}
+
 }  // namespace dataset
 }  // namespace mindspore

mindspore/ccsrc/minddata/dataset/kernels/image/soft_dvpp/soft_dvpp_decode_resize_jpeg_op.h

@@ -18,6 +18,7 @@
 
 #include <memory>
 #include <string>
+#include <vector>
 
 #include "minddata/dataset/core/tensor.h"
 #include "minddata/dataset/kernels/tensor_op.h"
@@ -34,6 +35,7 @@ class SoftDvppDecodeResizeJpegOp : public TensorOp {
   ~SoftDvppDecodeResizeJpegOp() = default;
 
   Status Compute(const std::shared_ptr<Tensor> &input, std::shared_ptr<Tensor> *output) override;
+  Status OutputShape(const std::vector<TensorShape> &inputs, std::vector<TensorShape> &outputs) override;
 
   std::string Name() const override { return kSoftDvppDecodeReiszeJpegOp; }
 

mindspore/dataset/transforms/vision/c_transforms.py

@@ -943,7 +943,7 @@ class SoftDvppDecodeResizeJpeg(cde.SoftDvppDecodeResizeJpegOp):
     @check_resize
     def __init__(self, size):
         if isinstance(size, int):
-            size = (size, size)
+            size = (size, 0)
         self.size = size
         super().__init__(*size)
 

tests/ut/python/dataset/test_soft_dvpp.py

@@ -84,7 +84,38 @@ def test_soft_dvpp_decode_random_crop_resize_jpeg(plot=False):
             visualize_image(image1, image2, mse)
         num_iter += 1
 
+def test_soft_dvpp_decode_resize_jpeg_supplement(plot=False):
+    """
+    Test SoftDvppDecodeResizeJpeg op
+    """
+    logger.info("test_random_decode_resize_op")
+
+    # First dataset
+    data1 = ds.TFRecordDataset(DATA_DIR, SCHEMA_DIR, columns_list=["image"], shuffle=False)
+    decode_op = vision.Decode()
+    resize_op = vision.Resize(256)
+    data1 = data1.map(input_columns=["image"], operations=[decode_op, resize_op])
+
+    # Second dataset
+    data2 = ds.TFRecordDataset(DATA_DIR, SCHEMA_DIR, columns_list=["image"], shuffle=False)
+    soft_dvpp_decode_resize_op = vision.SoftDvppDecodeResizeJpeg(256)
+    data2 = data2.map(input_columns=["image"], operations=soft_dvpp_decode_resize_op)
+
+    num_iter = 0
+    for item1, item2 in zip(data1.create_dict_iterator(), data2.create_dict_iterator()):
+        if num_iter > 0:
+            break
+        image1 = item1["image"]
+        image2 = item2["image"]
+        mse = diff_mse(image1, image2)
+        assert mse <= 0.02
+        logger.info("random_crop_decode_resize_op_{}, mse: {}".format(num_iter + 1, mse))
+        if plot:
+            visualize_image(image1, image2, mse)
+        num_iter += 1
 
 if __name__ == "__main__":
     test_soft_dvpp_decode_resize_jpeg(plot=True)
     test_soft_dvpp_decode_random_crop_resize_jpeg(plot=True)
+    test_soft_dvpp_decode_resize_jpeg_supplement(plot=True)
+    
\ No newline at end of file