Resolved some PR comments

Author: IasonTheodorou

.vscode/launch.json

@@ -6,7 +6,9 @@
             "type": "cppdbg",
             "request": "launch",
             "program": "/home/amigo/ros/noetic/system/devel/lib/sam_onnx_ros/test_sam_onnx_ros", // Path to the executable
-            "args": [], // Add any command-line arguments for your program here
+            "args": ["/home/amigo/Documents/repos/hero_sam.bak/sam_inference/model/SAM_encoder.onnx",
+                "/home/amigo/Documents/repos/hero_sam.bak/sam_inference/model/SAM_mask_decoder.onnx",
+                "/home/amigo/Documents/repos/hero_sam.bak/pipeline/build/images"], // Add any command-line arguments for your program here
             "stopAtEntry": false,
             "cwd": "${workspaceFolder}/build", // Set the working directory
             "environment": [],
@@ -19,7 +21,7 @@
                     "ignoreFailures": true
                 }
             ],
-            "preLaunchTask": "build-sam-project" // Ensure the project is built before debugging
+            // "preLaunchTask": "build-sam-project" // Ensure the project is built before debugging
         },
         {
             "name": "Debug GTest (pick binary)",

README.md

@@ -25,7 +25,7 @@ A high-performance C++ implementation for SAM (segment anything model) using ONN
 | Speed                       | 34ms       | 11.176ms | 5.984|
 | **Whole Pipeline (Enc+Dec)** |         | | |
 | Parameters                  | 9.66M     | -| -|
-| Su of Speed                       | 92.92ms      | - |-  |
+| Sum of Speed                       | 92.92ms      | - |-  |
 
 
 ## 📂 Project Structure

include/sam_onnx_ros/segmentation.hpp

@@ -16,8 +16,8 @@ Initialize(const std::filesystem::path& encoder_filename, const std::filesystem:
 
 void SegmentAnything(
     std::vector<std::unique_ptr<SAM>>& samSegmentors,
-    const SEG::_DL_INIT_PARAM& params_encoder,
-    const SEG::_DL_INIT_PARAM& params_decoder,
+    const SEG::DL_INIT_PARAM& params_encoder,
+    const SEG::DL_INIT_PARAM& params_decoder,
     const cv::Mat& img,
     std::vector<SEG::DL_RESULT>& resSam,
     SEG::DL_RESULT& res

src/main.cpp

@@ -34,7 +34,7 @@ int main(int argc, char *argv[])
 
             SegmentAnything(samSegmentors, params_encoder, params_decoder, img, resSam, res);
             #ifdef LOGGING
-            for (const auto& result : results)
+            for (const auto& result : resSam)
             {
                 std::cout << "Image path:   " << img_path << "\n"
                           << "# boxes:      " << result.boxes.size() << "\n"

src/sam_inference.cpp

@@ -61,7 +61,7 @@ const char* SAM::CreateSession(SEG::DL_INIT_PARAM& iParams)
     bool result = std::regex_search(iParams.modelPath, pattern);
     if (result)
     {
-        Ret = "[SAM]:Your model path is error. Change your model path without chinese characters.";
+        Ret = "[SAM]: Your model path is error. Change your model path without chinese characters.";
         CONSOLE_BRIDGE_logWarn("%s", Ret);
         return Ret;
     }
@@ -146,8 +146,8 @@ const char* SAM::RunSession(const cv::Mat& iImg, std::vector<SEG::DL_RESULT>& oR
         std::vector<int64_t> inputNodeDims;
         if (modelType == SEG::SAM_SEGMENT_ENCODER)
         {
-            // NCHW with H=imgSize[1], W=imgSize[0]  // FIX
-            inputNodeDims = { 1, 3, imgSize_.at(1), imgSize_.at(0) }; // FIX
+            // NCHW with H=imgSize[1], W=imgSize[0]
+            inputNodeDims = { 1, 3, imgSize_.at(1), imgSize_.at(0) };
         }
         else if (modelType == SEG::SAM_SEGMENT_DECODER)
         {
@@ -334,7 +334,7 @@ const char* SAM::TensorProcess_(clock_t& starttime_1, const cv::Mat& iImg,
             break;
         }
     default:
-        CONSOLE_BRIDGE_logError("[SAM]: " "Not support model type.");
+        CONSOLE_BRIDGE_logError("[SAM]: Unsupported model type.");
   }
 
   return RET_OK;
@@ -351,8 +351,8 @@ char* SAM::WarmUpSession_(SEG::MODEL_TYPE modelType)
     {
         float* blob = new float[iImg.total() * 3];
         utilities.BlobFromImage(processedImg, blob);
-        // NCHW: H=imgSize[1], W=imgSize[0]  // FIX
-        std::vector<int64_t> SAM_input_node_dims = { 1, 3, imgSize_.at(1), imgSize_.at(0) }; // FIX
+        // NCHW: H=imgSize[1], W=imgSize[0]
+        std::vector<int64_t> SAM_input_node_dims = { 1, 3, imgSize_.at(1), imgSize_.at(0) };
         switch (modelType)
         {
             case SEG::SAM_SEGMENT_ENCODER:

src/utils.cpp

@@ -28,13 +28,15 @@ char* Utils::PreProcess(const cv::Mat& iImg, std::vector<int> iImgSize, cv::Mat&
     {
         // Width-dominant: scale by target width (iImgSize[0])
         resizeScales_ = iImg.cols / static_cast<float>(iImgSize.at(0));
-        cv::resize(oImg, oImg, cv::Size(iImgSize.at(0), static_cast<int>(iImg.rows / resizeScales_))); // fixed
+        // Resize to target width, scaling height to maintain aspect ratio
+        cv::resize(oImg, oImg, cv::Size(iImgSize.at(0), static_cast<int>(iImg.rows / resizeScales_)));
     }
     else
     {
         // Height-dominant: scale by target height (iImgSize[1])
         resizeScales_ = iImg.rows / static_cast<float>(iImgSize.at(1));
-        cv::resize(oImg, oImg, cv::Size(static_cast<int>(iImg.cols / resizeScales_), iImgSize.at(1))); // fixed
+        // Resize width proportionally to target height to maintain aspect ratio (height-dominant case)
+        cv::resize(oImg, oImg, cv::Size(static_cast<int>(iImg.cols / resizeScales_), iImgSize.at(1)));
     }
 
     // Letterbox top-left into a canvas of size (H=iImgSize[1], W=iImgSize[0])