// ------------------------- OpenPose Library Tutorial - Pose - Example 1 - Extract from Image ------------------------- // This first example shows the user how to: // 1. Load an image (`filestream` module) // 2. Extract the pose of that image (`pose` module) // 3. Render the pose on a resized copy of the input image (`pose` module) // 4. Display the rendered pose (`gui` module) // In addition to the previous OpenPose modules, we also need to use: // 1. `core` module: for the Array class that the `pose` module needs // 2. `utilities` module: for the error & logging functions, i.e. op::error & op::log respectively // 3rdpary depencencies #include // DEFINE_bool, DEFINE_int32, DEFINE_int64, DEFINE_uint64, DEFINE_double, DEFINE_string #include // google::InitGoogleLogging, CHECK, CHECK_EQ, LOG, VLOG, ... // OpenPose dependencies #include #include #include #include #include // Gflags in the command line terminal. Check all the options by adding the flag `--help`, e.g. `rtpose.bin --help`. // Note: This command will show you flags for several files. Check only the flags for the file you are checking. E.g. for `rtpose`, look for `Flags from examples/openpose/rtpose.cpp:`. // Debugging DEFINE_int32(logging_level, 3, "The logging level. Integer in the range [0, 255]. 0 will output any log() message, while 255 will not output any." " Current OpenPose library messages are in the range 0-4: 1 for low priority messages and 4 for important ones."); // Producer DEFINE_string(image_path, "examples/media/COCO_val2014_000000000192.jpg", "Process the desired image."); // OpenPose DEFINE_string(model_pose, "COCO", "Model to be used (e.g. COCO, MPI, MPI_4_layers)."); DEFINE_string(model_folder, "models/", "Folder where the pose models (COCO and MPI) are located."); DEFINE_string(net_resolution, "656x368", "Multiples of 16."); DEFINE_string(resolution, "1280x720", "The image resolution (display). Use \"-1x-1\" to force the program to use the default images resolution."); DEFINE_int32(num_gpu_start, 0, "GPU device start number."); DEFINE_double(scale_gap, 0.3, "Scale gap between scales. No effect unless num_scales>1. Initial scale is always 1. If you want to change the initial scale, " "you actually want to multiply the `net_resolution` by your desired initial scale."); DEFINE_int32(num_scales, 1, "Number of scales to average."); // OpenPose Rendering DEFINE_double(alpha_pose, 0.6, "Blending factor (range 0-1) for the body part rendering. 1 will show it completely, 0 will hide it."); op::PoseModel gflagToPoseModel(const std::string& poseModeString) { op::log("", op::Priority::Low, __LINE__, __FUNCTION__, __FILE__); if (poseModeString == "COCO") return op::PoseModel::COCO_18; else if (poseModeString == "MPI") return op::PoseModel::MPI_15; else if (poseModeString == "MPI_4_layers") return op::PoseModel::MPI_15_4; else { op::error("String does not correspond to any model (COCO, MPI, MPI_4_layers)", __LINE__, __FUNCTION__, __FILE__); return op::PoseModel::COCO_18; } } // Google flags into program variables std::tuple gflagsToOpParameters() { op::log("", op::Priority::Low, __LINE__, __FUNCTION__, __FILE__); // outputSize cv::Size outputSize; auto nRead = sscanf(FLAGS_resolution.c_str(), "%dx%d", &outputSize.width, &outputSize.height); op::checkE(nRead, 2, "Error, resolution format (" + FLAGS_resolution + ") invalid, should be e.g., 960x540 ", __LINE__, __FUNCTION__, __FILE__); // netInputSize cv::Size netInputSize; nRead = sscanf(FLAGS_net_resolution.c_str(), "%dx%d", &netInputSize.width, &netInputSize.height); op::checkE(nRead, 2, "Error, net resolution format (" + FLAGS_net_resolution + ") invalid, should be e.g., 656x368 (multiples of 16)", __LINE__, __FUNCTION__, __FILE__); // netOutputSize const auto netOutputSize = netInputSize; // poseModel const auto poseModel = gflagToPoseModel(FLAGS_model_pose); // Check no contradictory flags enabled if (FLAGS_alpha_pose < 0. || FLAGS_alpha_pose > 1.) op::error("Alpha value for blending must be in the range [0,1].", __LINE__, __FUNCTION__, __FILE__); if (FLAGS_scale_gap <= 0. && FLAGS_num_scales > 1) op::error("Uncompatible flag configuration: scale_gap must be greater than 0 or num_scales = 1.", __LINE__, __FUNCTION__, __FILE__); // Logging and return result op::log("", op::Priority::Low, __LINE__, __FUNCTION__, __FILE__); return std::make_tuple(outputSize, netInputSize, netOutputSize, poseModel); } int openPoseTutorialPose1() { op::log("OpenPose Library Tutorial - Example 1.", op::Priority::Max); // ------------------------- INITIALIZATION ------------------------- // Step 1 - Set logging level // - 0 will output all the logging messages // - 255 will output nothing op::check(0 <= FLAGS_logging_level && FLAGS_logging_level <= 255, "Wrong logging_level value.", __LINE__, __FUNCTION__, __FILE__); op::ConfigureLog::setPriorityThreshold((op::Priority)FLAGS_logging_level); // Step 2 - Read Google flags (user defined configuration) cv::Size outputSize; cv::Size netInputSize; cv::Size netOutputSize; op::PoseModel poseModel; std::tie(outputSize, netInputSize, netOutputSize, poseModel) = gflagsToOpParameters(); // Step 3 - Initialize all required classes op::CvMatToOpInput cvMatToOpInput{netInputSize, FLAGS_num_scales, (float)FLAGS_scale_gap}; op::CvMatToOpOutput cvMatToOpOutput{outputSize}; op::PoseExtractorCaffe poseExtractorCaffe{netInputSize, netOutputSize, outputSize, FLAGS_num_scales, (float)FLAGS_scale_gap, poseModel, FLAGS_model_folder, FLAGS_num_gpu_start}; op::PoseRenderer poseRenderer{netOutputSize, outputSize, poseModel, nullptr, (float)FLAGS_alpha_pose}; op::OpOutputToCvMat opOutputToCvMat{outputSize}; const cv::Size windowedSize = outputSize; op::FrameDisplayer frameDisplayer{windowedSize, "OpenPose Tutorial - Example 1"}; // Step 4 - Initialize resources on desired thread (in this case single thread, i.e. we init resources here) poseExtractorCaffe.initializationOnThread(); poseRenderer.initializationOnThread(); // ------------------------- POSE ESTIMATION AND RENDERING ------------------------- // Step 1 - Read and load image, error if empty (possibly wrong path) cv::Mat inputImage = op::loadImage(FLAGS_image_path, CV_LOAD_IMAGE_COLOR); // Alternative: cv::imread(FLAGS_image_path, CV_LOAD_IMAGE_COLOR); if(inputImage.empty()) op::error("Could not open or find the image: " + FLAGS_image_path, __LINE__, __FUNCTION__, __FILE__); // Step 2 - Format input image to OpenPose input and output formats const auto netInputArray = cvMatToOpInput.format(inputImage); double scaleInputToOutput; op::Array outputArray; std::tie(scaleInputToOutput, outputArray) = cvMatToOpOutput.format(inputImage); // Step 3 - Estimate poseKeyPoints poseExtractorCaffe.forwardPass(netInputArray, inputImage.size()); const auto poseKeyPoints = poseExtractorCaffe.getPoseKeyPoints(); // Step 4 - Render poseKeyPoints poseRenderer.renderPose(outputArray, poseKeyPoints); // Step 5 - OpenPose output format to cv::Mat auto outputImage = opOutputToCvMat.formatToCvMat(outputArray); // ------------------------- SHOWING RESULT AND CLOSING ------------------------- // Step 1 - Show results frameDisplayer.displayFrame(outputImage, 0); // Alternative: cv::imshow(outputImage) + cv::waitKey(0) // Step 2 - Logging information message op::log("Example 1 successfully finished.", op::Priority::Max); // Return successful message return 0; } int main(int argc, char *argv[]) { // Initializing google logging (Caffe uses it for logging) google::InitGoogleLogging("openPoseTutorialPose1"); // Parsing command line flags gflags::ParseCommandLineFlags(&argc, &argv, true); // Running openPoseTutorialPose1 return openPoseTutorialPose1(); }