Remove uncoditional texture params reset to make float textures handled correctly.

PiperOrigin-RevId: 564869245

mediapipe/calculators/image/image_transformation_calculator.cc

@@ -660,6 +660,7 @@ absl::Status ImageTransformationCalculator::RenderGpu(CalculatorContext* cc) {
   glBindTexture(src1.target(), src1.name());
 
   if (interpolation_mode_ == ImageTransformationCalculatorOptions::NEAREST) {
+    // TODO: revert texture params.
     glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
     glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
   }
@@ -669,10 +670,6 @@ absl::Status ImageTransformationCalculator::RenderGpu(CalculatorContext* cc) {
       rotation, flip_horizontally_, flip_vertically_,
       /*flip_texture=*/false));
 
-  // Reset interpolation modes to MediaPipe defaults.
-  glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
-  glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
-
   glActiveTexture(GL_TEXTURE1);
   glBindTexture(src1.target(), 0);
 

mediapipe/calculators/image/image_transformation_calculator_test.cc

@@ -18,6 +18,7 @@
 #include "mediapipe/framework/port/parse_text_proto.h"
 #include "testing/base/public/gmock.h"
 #include "testing/base/public/googletest.h"
+#include "third_party/OpenCV/core.hpp"  // IWYU pragma: keep
 #include "third_party/OpenCV/core/mat.hpp"
 
 namespace mediapipe {
@@ -93,6 +94,67 @@ TEST(ImageTransformationCalculatorTest, NearestNeighborResizing) {
   }
 }
 
+TEST(ImageTransformationCalculatorTest,
+     NearestNeighborResizingWorksForFloatInput) {
+  cv::Mat input_mat;
+  cv::cvtColor(cv::imread(file::JoinPath("./",
+                                         "/mediapipe/calculators/"
+                                         "image/testdata/binary_mask.png")),
+               input_mat, cv::COLOR_BGR2GRAY);
+  Packet input_image_packet = MakePacket<ImageFrame>(
+      ImageFormat::VEC32F1, input_mat.size().width, input_mat.size().height);
+  cv::Mat packet_mat_view =
+      formats::MatView(&(input_image_packet.Get<ImageFrame>()));
+  input_mat.convertTo(packet_mat_view, CV_32FC1, 1 / 255.f);
+
+  std::vector<std::pair<int, int>> output_dims{
+      {256, 333}, {512, 512}, {1024, 1024}};
+
+  for (auto& output_dim : output_dims) {
+    Packet input_output_dim_packet =
+        MakePacket<std::pair<int, int>>(output_dim);
+    std::vector<std::string> scale_modes{"FIT", "STRETCH"};
+    for (const auto& scale_mode : scale_modes) {
+      CalculatorGraphConfig::Node node_config =
+          ParseTextProtoOrDie<CalculatorGraphConfig::Node>(
+              absl::Substitute(R"(
+          calculator: "ImageTransformationCalculator"
+          input_stream: "IMAGE:input_image"
+          input_stream: "OUTPUT_DIMENSIONS:image_size"
+          output_stream: "IMAGE:output_image"
+          options: {
+            [mediapipe.ImageTransformationCalculatorOptions.ext]: {
+              scale_mode: $0
+              interpolation_mode: NEAREST
+            }
+          })",
+                               scale_mode));
+
+      CalculatorRunner runner(node_config);
+      runner.MutableInputs()->Tag("IMAGE").packets.push_back(
+          input_image_packet.At(Timestamp(0)));
+      runner.MutableInputs()
+          ->Tag("OUTPUT_DIMENSIONS")
+          .packets.push_back(input_output_dim_packet.At(Timestamp(0)));
+
+      MP_ASSERT_OK(runner.Run());
+      const auto& outputs = runner.Outputs();
+      ASSERT_EQ(outputs.NumEntries(), 1);
+      const std::vector<Packet>& packets = outputs.Tag("IMAGE").packets;
+      ASSERT_EQ(packets.size(), 1);
+      const auto& result = packets[0].Get<ImageFrame>();
+      ASSERT_EQ(output_dim.first, result.Width());
+      ASSERT_EQ(output_dim.second, result.Height());
+
+      auto unique_input_values = computeUniqueValues(packet_mat_view);
+      auto unique_output_values =
+          computeUniqueValues(formats::MatView(&result));
+      EXPECT_THAT(unique_input_values,
+                  ::testing::ContainerEq(unique_output_values));
+    }
+  }
+}
+
 TEST(ImageTransformationCalculatorTest, NearestNeighborResizingGpu) {
   cv::Mat input_mat;
   cv::cvtColor(cv::imread(file::JoinPath("./",
@@ -170,5 +232,84 @@ TEST(ImageTransformationCalculatorTest, NearestNeighborResizingGpu) {
   }
 }
 
+TEST(ImageTransformationCalculatorTest,
+     NearestNeighborResizingWorksForFloatTexture) {
+  cv::Mat input_mat;
+  cv::cvtColor(cv::imread(file::JoinPath("./",
+                                         "/mediapipe/calculators/"
+                                         "image/testdata/binary_mask.png")),
+               input_mat, cv::COLOR_BGR2GRAY);
+  Packet input_image_packet = MakePacket<ImageFrame>(
+      ImageFormat::VEC32F1, input_mat.size().width, input_mat.size().height);
+  cv::Mat packet_mat_view =
+      formats::MatView(&(input_image_packet.Get<ImageFrame>()));
+  input_mat.convertTo(packet_mat_view, CV_32FC1, 1 / 255.f);
+
+  std::vector<std::pair<int, int>> output_dims{
+      {256, 333}, {512, 512}, {1024, 1024}};
+
+  for (auto& output_dim : output_dims) {
+    std::vector<std::string> scale_modes{"FIT"};  //, "STRETCH"};
+    for (const auto& scale_mode : scale_modes) {
+      CalculatorGraphConfig graph_config =
+          ParseTextProtoOrDie<CalculatorGraphConfig>(
+              absl::Substitute(R"(
+          input_stream: "input_image"
+          input_stream: "image_size"
+          output_stream: "output_image"
+
+          node {
+            calculator: "ImageFrameToGpuBufferCalculator"
+            input_stream: "input_image"
+            output_stream: "input_image_gpu"
+          }
+
+          node {
+            calculator: "ImageTransformationCalculator"
+            input_stream: "IMAGE_GPU:input_image_gpu"
+            input_stream: "OUTPUT_DIMENSIONS:image_size"
+            output_stream: "IMAGE_GPU:output_image_gpu"
+            options: {
+              [mediapipe.ImageTransformationCalculatorOptions.ext]: {
+                scale_mode: $0
+                interpolation_mode: NEAREST
+              }
+            }
+          }
+          node {
+            calculator: "GpuBufferToImageFrameCalculator"
+            input_stream: "output_image_gpu"
+            output_stream: "output_image"
+          })",
+                               scale_mode));
+
+      std::vector<Packet> output_image_packets;
+      tool::AddVectorSink("output_image", &graph_config, &output_image_packets);
+
+      CalculatorGraph graph(graph_config);
+      MP_ASSERT_OK(graph.StartRun({}));
+
+      MP_ASSERT_OK(graph.AddPacketToInputStream(
+          "input_image", input_image_packet.At(Timestamp(0))));
+      MP_ASSERT_OK(graph.AddPacketToInputStream(
+          "image_size",
+          MakePacket<std::pair<int, int>>(output_dim).At(Timestamp(0))));
+
+      MP_ASSERT_OK(graph.WaitUntilIdle());
+      ASSERT_THAT(output_image_packets, testing::SizeIs(1));
+
+      const auto& output_image = output_image_packets[0].Get<ImageFrame>();
+      ASSERT_EQ(output_dim.first, output_image.Width());
+      ASSERT_EQ(output_dim.second, output_image.Height());
+
+      auto unique_input_values = computeUniqueValues(packet_mat_view);
+      auto unique_output_values =
+          computeUniqueValues(formats::MatView(&output_image));
+      EXPECT_THAT(unique_input_values,
+                  ::testing::ContainerEq(unique_output_values));
+    }
+  }
+}
+
 }  // namespace
 }  // namespace mediapipe