mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-08-06 14:36:36 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #26486 from gursimarsingh:object_detection_engine_update

Browse Source 
Code Fixes and changed post processing based on models.yml in Object Detection Sample #26486

## Major Changes

1. Changes to add findModel support for config file in models like yolov4, yolov4-tiny, yolov3, ssd_caffe, tiny-yolo-voc, ssd_tf and faster_rcnn_tf.
2. Added new model and config download links for ssd_caffe, as previous links were not working.
3. Switched to DNN ENGINE_CLASSIC for non-cpu convig as new engine does not support it.
4. Fixes in python sample related to yolov5 usage.

### Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
      Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
This commit is contained in:
Gursimar Singh 2024-12-02 11:35:25 +05:30 committed by GitHub
parent b476ed6d06
commit efbe580ff3
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
5 changed files with 110 additions and 74 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
samples/dnn/common.hpp
View File

@ -62,7 +62,12 @@ std::string genArgument(const std::string& argName, const std::string& help,
FileNode value = node[argName]; FileNode value = node[argName];
if (argName.find("sha1") != std::string::npos) { if (argName.find("sha1") != std::string::npos) {
std::string prefix = argName.substr(0, argName.find("sha1")); std::string prefix = argName.substr(0, argName.find("sha1"));
value = node[prefix+"load_info"][argName]; if (prefix == "config_"){
value = node[prefix+"load_info"]["sha1"];
}
else{
value = node[prefix+"load_info"][argName];
}
} }
if (argName.find("download_sha") != std::string::npos) { if (argName.find("download_sha") != std::string::npos) {
std::string prefix = argName.substr(0, argName.find("download_sha")); std::string prefix = argName.substr(0, argName.find("download_sha"));
@ -183,6 +188,8 @@ std::string genPreprocArguments(const std::string& modelName, const std::string&
modelName, zooFile)+ modelName, zooFile)+
genArgument(prefix + "sha1", "Optional path to hashsum of downloaded model to be loaded from models.yml", genArgument(prefix + "sha1", "Optional path to hashsum of downloaded model to be loaded from models.yml",
modelName, zooFile)+ modelName, zooFile)+
genArgument(prefix + "config_sha1", "Optional path to hashsum of downloaded config to be loaded from models.yml",
modelName, zooFile)+
genArgument(prefix + "download_sha", "Optional path to hashsum of downloaded model to be loaded from models.yml", genArgument(prefix + "download_sha", "Optional path to hashsum of downloaded model to be loaded from models.yml",
modelName, zooFile); modelName, zooFile);
} }

7
samples/dnn/common.py
View File

@ -19,7 +19,10 @@ def add_argument(zoo, parser, name, help, required=False, default=None, type=Non
if "sha1" in name: if "sha1" in name:
prefix = name.replace("sha1", "") prefix = name.replace("sha1", "")
value = node.getNode(prefix + "load_info") value = node.getNode(prefix + "load_info")
value = value.getNode(name) if prefix == "config_":
value = value.getNode("sha1")
else:
value = value.getNode(name)
if "download_sha" in name: if "download_sha" in name:
prefix = name.replace("download_sha", "") prefix = name.replace("download_sha", "")
value = node.getNode(prefix + "load_info") value = node.getNode(prefix + "load_info")
@ -97,6 +100,8 @@ def add_preproc_args(zoo, parser, sample, alias=None, prefix=""):
help='An index of background class in predictions. If not negative, exclude such class from list of classes.', alias=alias) help='An index of background class in predictions. If not negative, exclude such class from list of classes.', alias=alias)
add_argument(zoo, parser, prefix+'sha1', type=str, add_argument(zoo, parser, prefix+'sha1', type=str,
help='Optional path to hashsum of downloaded model to be loaded from models.yml', alias=alias) help='Optional path to hashsum of downloaded model to be loaded from models.yml', alias=alias)
add_argument(zoo, parser, prefix+'config_sha1', type=str,
help='Optional path to hashsum of downloaded config to be loaded from models.yml', alias=alias)
add_argument(zoo, parser, prefix+'download_sha', type=str, add_argument(zoo, parser, prefix+'download_sha', type=str,
help='Optional path to hashsum of downloaded model to be loaded from models.yml', alias=alias) help='Optional path to hashsum of downloaded model to be loaded from models.yml', alias=alias)

18
samples/dnn/models.yml
View File

@ -89,8 +89,8 @@ yolov5l:
width: 640 width: 640
height: 640 height: 640
rgb: true rgb: true
classes: "object_detection_classes_yolo.txt" labels: "object_detection_classes_yolo.txt"
background_label_id: 0 postprocessing: "yolov5"
sample: "object_detection" sample: "object_detection"
# YOLO4 object detection family from Darknet (https://github.com/AlexeyAB/darknet) # YOLO4 object detection family from Darknet (https://github.com/AlexeyAB/darknet)
@ -112,6 +112,7 @@ yolov4:
rgb: true rgb: true
labels: "object_detection_classes_yolo.txt" labels: "object_detection_classes_yolo.txt"
background_label_id: 0 background_label_id: 0
postprocessing: "darknet"
sample: "object_detection" sample: "object_detection"
yolov4-tiny: yolov4-tiny:
@ -130,6 +131,7 @@ yolov4-tiny:
rgb: true rgb: true
labels: "object_detection_classes_yolo.txt" labels: "object_detection_classes_yolo.txt"
background_label_id: 0 background_label_id: 0
postprocessing: "darknet"
sample: "object_detection" sample: "object_detection"
yolov3: yolov3:
@ -148,6 +150,7 @@ yolov3:
rgb: true rgb: true
labels: "object_detection_classes_yolo.txt" labels: "object_detection_classes_yolo.txt"
background_label_id: 0 background_label_id: 0
postprocessing: "darknet"
sample: "object_detection" sample: "object_detection"
tiny-yolo-voc: tiny-yolo-voc:
@ -166,14 +169,18 @@ tiny-yolo-voc:
rgb: true rgb: true
labels: "object_detection_classes_pascal_voc.txt" labels: "object_detection_classes_pascal_voc.txt"
background_label_id: 0 background_label_id: 0
postprocessing: "darknet"
sample: "object_detection" sample: "object_detection"
# Caffe implementation of SSD model from https://github.com/chuanqi305/MobileNet-SSD # Caffe implementation of SSD model from https://github.com/PINTO0309/MobileNet-SSD-RealSense
ssd_caffe: ssd_caffe:
load_info: load_info:
url: "https://drive.google.com/uc?export=download&id=0B3gersZ2cHIxRm5PMWRoTkdHdHc" url: "https://github.com/PINTO0309/MobileNet-SSD-RealSense/raw/refs/heads/master/caffemodel/MobileNetSSD/MobileNetSSD_deploy.caffemodel"
sha1: "994d30a8afaa9e754d17d2373b2d62a7dfbaaf7a" sha1: "994d30a8afaa9e754d17d2373b2d62a7dfbaaf7a"
model: "MobileNetSSD_deploy.caffemodel" model: "MobileNetSSD_deploy.caffemodel"
config_load_info:
url: "https://github.com/PINTO0309/MobileNet-SSD-RealSense/raw/refs/heads/master/caffemodel/MobileNetSSD/MobileNetSSD_deploy.prototxt"
sha1: "25c8404cecdef638c2bd9ac7f3b46a8b96897deb"
config: "MobileNetSSD_deploy.prototxt" config: "MobileNetSSD_deploy.prototxt"
mean: [127.5, 127.5, 127.5] mean: [127.5, 127.5, 127.5]
scale: 0.007843 scale: 0.007843
@ -181,6 +188,7 @@ ssd_caffe:
height: 300 height: 300
rgb: false rgb: false
labels: "object_detection_classes_pascal_voc.txt" labels: "object_detection_classes_pascal_voc.txt"
postprocessing: "ssd"
sample: "object_detection" sample: "object_detection"
# TensorFlow implementation of SSD model from https://github.com/tensorflow/models/tree/master/research/object_detection # TensorFlow implementation of SSD model from https://github.com/tensorflow/models/tree/master/research/object_detection
@ -202,6 +210,7 @@ ssd_tf:
height: 300 height: 300
rgb: true rgb: true
labels: "object_detection_classes_coco.txt" labels: "object_detection_classes_coco.txt"
postprocessing: "ssd"
sample: "object_detection" sample: "object_detection"
# TensorFlow implementation of Faster-RCNN model from https://github.com/tensorflow/models/tree/master/research/object_detection # TensorFlow implementation of Faster-RCNN model from https://github.com/tensorflow/models/tree/master/research/object_detection
@ -222,6 +231,7 @@ faster_rcnn_tf:
width: 800 width: 800
height: 600 height: 600
rgb: true rgb: true
postprocessing: "ssd"
sample: "object_detection" sample: "object_detection"
################################################################################ ################################################################################

65
samples/dnn/object_detection.cpp
View File

@ -76,7 +76,7 @@ string modelName, framework;
static void preprocess(const Mat& frame, Net& net, Size inpSize); static void preprocess(const Mat& frame, Net& net, Size inpSize);
static void postprocess(Mat& frame, const vector<Mat>& outs, Net& net, int backend, vector<int>& classIds, vector<float>& confidences, vector<Rect>& boxes, const string yolo_name); static void postprocess(Mat& frame, const vector<Mat>& outs, Net& net, int backend, vector<int>& classIds, vector<float>& confidences, vector<Rect>& boxes, const string postprocessing);
static void drawPred(vector<int>& classIds, vector<float>& confidences, vector<Rect>& boxes, Mat& frame, FontFace& sans, int stdSize, int stdWeight, int stdImgSize, int stdThickness); static void drawPred(vector<int>& classIds, vector<float>& confidences, vector<Rect>& boxes, Mat& frame, FontFace& sans, int stdSize, int stdWeight, int stdImgSize, int stdThickness);
@ -91,7 +91,7 @@ static void yoloPostProcessing(
vector<Rect2d>& keep_boxes, vector<Rect2d>& keep_boxes,
float conf_threshold, float conf_threshold,
float iou_threshold, float iou_threshold,
const string& yolo_name); const string& postprocessing);
static void printAliases(string& zooFile){ static void printAliases(string& zooFile){
vector<string> aliases = findAliases(zooFile, "object_detection"); vector<string> aliases = findAliases(zooFile, "object_detection");
@ -195,12 +195,13 @@ int main(int argc, char** argv)
inpHeight = parser.get<int>("height"); inpHeight = parser.get<int>("height");
int async = parser.get<int>("async"); int async = parser.get<int>("async");
paddingValue = parser.get<float>("padvalue"); paddingValue = parser.get<float>("padvalue");
const string yolo_name = parser.get<String>("postprocessing"); const string postprocessing = parser.get<String>("postprocessing");
paddingMode = static_cast<ImagePaddingMode>(parser.get<int>("paddingmode")); paddingMode = static_cast<ImagePaddingMode>(parser.get<int>("paddingmode"));
//![preprocess_params] //![preprocess_params]
String sha1 = parser.get<String>("sha1"); String sha1 = parser.get<String>("sha1");
String config_sha1 = parser.get<String>("config_sha1");
const string modelPath = findModel(parser.get<String>("model"), sha1); const string modelPath = findModel(parser.get<String>("model"), sha1);
const string configPath = findFile(parser.get<String>("config")); const string configPath = findModel(parser.get<String>("config"), config_sha1);
framework = modelPath.substr(modelPath.rfind('.') + 1); framework = modelPath.substr(modelPath.rfind('.') + 1);
if (parser.has("labels")) if (parser.has("labels"))
@ -216,7 +217,11 @@ int main(int argc, char** argv)
} }
} }
//![read_net] //![read_net]
Net net = readNet(modelPath, configPath); EngineType engine = ENGINE_AUTO;
if ((parser.get<String>("backend") != "default") || (parser.get<String>("target") != "cpu")){
engine = ENGINE_CLASSIC;
}
Net net = readNet(modelPath, configPath, "", engine);
int backend = getBackendID(parser.get<String>("backend")); int backend = getBackendID(parser.get<String>("backend"));
net.setPreferableBackend(backend); net.setPreferableBackend(backend);
net.setPreferableTarget(getTargetID(parser.get<String>("target"))); net.setPreferableTarget(getTargetID(parser.get<String>("target")));
@ -230,7 +235,7 @@ int main(int argc, char** argv)
// Open a video file or an image file or a camera stream. // Open a video file or an image file or a camera stream.
VideoCapture cap; VideoCapture cap;
bool openSuccess = parser.has("input") ? cap.open(parser.get<String>("input")) : cap.open(parser.get<int>("device")); bool openSuccess = parser.has("input") ? cap.open(findFile(parser.get<String>("input"))) : cap.open(parser.get<int>("device"));
if (!openSuccess){ if (!openSuccess){
cout << "Could not open input file or camera device" << endl; cout << "Could not open input file or camera device" << endl;
return 0; return 0;
@ -324,7 +329,7 @@ int main(int argc, char** argv)
classIds.clear(); classIds.clear();
confidences.clear(); confidences.clear();
boxes.clear(); boxes.clear();
postprocess(frame, outs, net, backend, classIds, confidences, boxes, yolo_name); postprocess(frame, outs, net, backend, classIds, confidences, boxes, postprocessing);
drawPred(classIds, confidences, boxes, frame, sans, stdSize, stdWeight, stdImgSize, stdThickness); drawPred(classIds, confidences, boxes, frame, sans, stdSize, stdWeight, stdImgSize, stdThickness);
@ -354,7 +359,7 @@ int main(int argc, char** argv)
CV_Error(Error::StsNotImplemented, "Asynchronous forward is supported only with Inference Engine backend."); CV_Error(Error::StsNotImplemented, "Asynchronous forward is supported only with Inference Engine backend.");
// Threading is disabled, run synchronously // Threading is disabled, run synchronously
Mat frame, blob; Mat frame, blob;
while (waitKey(100) < 0) { while (waitKey(1) < 0) {
cap >> frame; cap >> frame;
if (frame.empty()) { if (frame.empty()) {
waitKey(); waitKey();
@ -369,7 +374,7 @@ int main(int argc, char** argv)
confidences.clear(); confidences.clear();
boxes.clear(); boxes.clear();
postprocess(frame, outs, net, backend, classIds, confidences, boxes, yolo_name); postprocess(frame, outs, net, backend, classIds, confidences, boxes, postprocessing);
drawPred(classIds, confidences, boxes, frame, sans, stdSize, stdWeight, stdImgSize, stdThickness); drawPred(classIds, confidences, boxes, frame, sans, stdSize, stdWeight, stdImgSize, stdThickness);
@ -379,7 +384,7 @@ int main(int argc, char** argv)
int weight = static_cast<int>((stdWeight * imgWidth) / (stdImgSize * 1.5)); int weight = static_cast<int>((stdWeight * imgWidth) / (stdImgSize * 1.5));
double freq = getTickFrequency() / 1000; double freq = getTickFrequency() / 1000;
double t = net.getPerfProfile(layersTimes) / freq; double t = net.getPerfProfile(layersTimes) / freq;
string label = format("Inference time: %.2f ms", t); string label = format("FPS: %.2f", 1000/t);
putText(frame, label, Point(0, size), Scalar(0, 255, 0), sans, size, weight); putText(frame, label, Point(0, size), Scalar(0, 255, 0), sans, size, weight);
imshow(kWinName, frame); imshow(kWinName, frame);
} }
@ -414,15 +419,6 @@ void preprocess(const Mat& frame, Net& net, Size inpSize)
// Set the blob as the network input // Set the blob as the network input
net.setInput(inp); net.setInput(inp);
// Check if the model is Faster-RCNN or R-FCN
if (net.getLayer(0)->outputNameToIndex("im_info") != -1)
{
// Resize the frame and prepare imInfo
resize(frame, frame, size);
Mat imInfo = (Mat_<float>(1, 3) << size.height, size.width, 1.6f);
net.setInput(imInfo, "im_info");
}
} }
void yoloPostProcessing( void yoloPostProcessing(
@ -432,7 +428,7 @@ void yoloPostProcessing(
vector<Rect2d>& keep_boxes, vector<Rect2d>& keep_boxes,
float conf_threshold, float conf_threshold,
float iou_threshold, float iou_threshold,
const string& yolo_name) const string& postprocessing)
{ {
// Retrieve // Retrieve
vector<int> classIds; vector<int> classIds;
@ -441,12 +437,12 @@ void yoloPostProcessing(
vector<Mat> outs_copy = outs; vector<Mat> outs_copy = outs;
if (yolo_name == "yolov8") if (postprocessing == "yolov8")
{ {
transposeND(outs_copy[0], {0, 2, 1}, outs_copy[0]); transposeND(outs_copy[0], {0, 2, 1}, outs_copy[0]);
} }
if (yolo_name == "yolonas") if (postprocessing == "yolonas")
{ {
// outs contains 2 elements of shape [1, 8400, 80] and [1, 8400, 4]. Concat them to get [1, 8400, 84] // outs contains 2 elements of shape [1, 8400, 80] and [1, 8400, 4]. Concat them to get [1, 8400, 84]
Mat concat_out; Mat concat_out;
@ -467,16 +463,16 @@ void yoloPostProcessing(
for (int i = 0; i < preds.rows; ++i) for (int i = 0; i < preds.rows; ++i)
{ {
// filter out non-object // filter out non-object
float obj_conf = (yolo_name == "yolov8" || yolo_name == "yolonas") ? 1.0f : preds.at<float>(i, 4); float obj_conf = (postprocessing == "yolov8" || postprocessing == "yolonas") ? 1.0f : preds.at<float>(i, 4);
if (obj_conf < conf_threshold) if (obj_conf < conf_threshold)
continue; continue;
Mat scores = preds.row(i).colRange((yolo_name == "yolov8" || yolo_name == "yolonas") ? 4 : 5, preds.cols); Mat scores = preds.row(i).colRange((postprocessing == "yolov8" || postprocessing == "yolonas") ? 4 : 5, preds.cols);
double conf; double conf;
Point maxLoc; Point maxLoc;
minMaxLoc(scores, 0, &conf, 0, &maxLoc); minMaxLoc(scores, 0, &conf, 0, &maxLoc);
conf = (yolo_name == "yolov8" || yolo_name == "yolonas") ? conf : conf * obj_conf; conf = (postprocessing == "yolov8" || postprocessing == "yolonas") ? conf : conf * obj_conf;
if (conf < conf_threshold) if (conf < conf_threshold)
continue; continue;
@ -488,7 +484,7 @@ void yoloPostProcessing(
double h = det[3]; double h = det[3];
// [x1, y1, x2, y2] // [x1, y1, x2, y2]
if (yolo_name == "yolonas") { if (postprocessing == "yolonas") {
boxes.push_back(Rect2d(cx, cy, w, h)); boxes.push_back(Rect2d(cx, cy, w, h));
} else { } else {
boxes.push_back(Rect2d(cx - 0.5 * w, cy - 0.5 * h, boxes.push_back(Rect2d(cx - 0.5 * w, cy - 0.5 * h,
@ -511,12 +507,10 @@ void yoloPostProcessing(
} }
} }
void postprocess(Mat& frame, const vector<Mat>& outs, Net& net, int backend, vector<int>& classIds, vector<float>& confidences, vector<Rect>& boxes, const string yolo_name) void postprocess(Mat& frame, const vector<Mat>& outs, Net& net, int backend, vector<int>& classIds, vector<float>& confidences, vector<Rect>& boxes, const string postprocessing)
{ {
static vector<int> outLayers = net.getUnconnectedOutLayers(); static vector<int> outLayers = net.getUnconnectedOutLayers();
static string outLayerType = net.getLayer(outLayers[0])->type; if (postprocessing == "ssd")
if (outLayerType == "DetectionOutput")
{ {
// Network produces output blob with a shape 1x1xNx7 where N is a number of // Network produces output blob with a shape 1x1xNx7 where N is a number of
// detections and an every detection is a vector of values // detections and an every detection is a vector of values
@ -552,7 +546,7 @@ void postprocess(Mat& frame, const vector<Mat>& outs, Net& net, int backend, vec
} }
} }
} }
else if (outLayerType == "Region") else if (postprocessing == "darknet")
{ {
for (size_t i = 0; i < outs.size(); ++i) for (size_t i = 0; i < outs.size(); ++i)
{ {
@ -582,7 +576,7 @@ void postprocess(Mat& frame, const vector<Mat>& outs, Net& net, int backend, vec
} }
} }
} }
else if (outLayerType == "Identity") else if (postprocessing == "yolov8" || postprocessing == "yolov5")
{ {
//![forward_buffers] //![forward_buffers]
vector<int> keep_classIds; vector<int> keep_classIds;
@ -591,7 +585,7 @@ void postprocess(Mat& frame, const vector<Mat>& outs, Net& net, int backend, vec
//![forward_buffers] //![forward_buffers]
//![postprocess] //![postprocess]
yoloPostProcessing(outs, keep_classIds, keep_confidences, keep_boxes, confThreshold, nmsThreshold, yolo_name); yoloPostProcessing(outs, keep_classIds, keep_confidences, keep_boxes, confThreshold, nmsThreshold, postprocessing);
//![postprocess] //![postprocess]
for (size_t i = 0; i < keep_classIds.size(); ++i) for (size_t i = 0; i < keep_classIds.size(); ++i)
@ -614,12 +608,13 @@ void postprocess(Mat& frame, const vector<Mat>& outs, Net& net, int backend, vec
} }
else else
{ {
CV_Error(Error::StsNotImplemented, "Unknown output layer type: " + outLayerType); cout<< ("Unknown postprocessing method: " + postprocessing)<<endl;
exit(-1);
} }
// NMS is used inside Region layer only on DNN_BACKEND_OPENCV for other backends we need NMS in sample // NMS is used inside Region layer only on DNN_BACKEND_OPENCV for other backends we need NMS in sample
// or NMS is required if the number of outputs > 1 // or NMS is required if the number of outputs > 1
if (outLayers.size() > 1 || (outLayerType == "Region" && backend != DNN_BACKEND_OPENCV)) if (outLayers.size() > 1 || (postprocessing == "darknet" && backend != DNN_BACKEND_OPENCV))
{ {
map<int, vector<size_t> > class2indices; map<int, vector<size_t> > class2indices;
for (size_t i = 0; i < classIds.size(); i++) for (size_t i = 0; i < classIds.size(); i++)

85
samples/dnn/object_detection.py
View File

@ -37,9 +37,6 @@ parser.add_argument('--out_tf_graph', default='graph.pbtxt',
help='For models from TensorFlow Object Detection API, you may ' help='For models from TensorFlow Object Detection API, you may '
'pass a .config file which was used for training through --config ' 'pass a .config file which was used for training through --config '
'argument. This way an additional .pbtxt file with TensorFlow graph will be created.') 'argument. This way an additional .pbtxt file with TensorFlow graph will be created.')
parser.add_argument('--framework', choices=['caffe', 'tensorflow', 'darknet', 'dldt', 'onnx'],
help='Optional name of an origin framework of the model. '
'Detect it automatically if it does not set.')
parser.add_argument('--thr', type=float, default=0.5, help='Confidence threshold') parser.add_argument('--thr', type=float, default=0.5, help='Confidence threshold')
parser.add_argument('--nms', type=float, default=0.4, help='Non-maximum suppression threshold') parser.add_argument('--nms', type=float, default=0.4, help='Non-maximum suppression threshold')
parser.add_argument('--backend', default="default", type=str, choices=backends, parser.add_argument('--backend', default="default", type=str, choices=backends,
@ -76,8 +73,9 @@ if args.alias is None or hasattr(args, 'help'):
args.model = findModel(args.model, args.sha1) args.model = findModel(args.model, args.sha1)
if args.config is not None: if args.config is not None:
args.config = findFile(args.config) args.config = findModel(args.config, args.config_sha1)
args.labels = findFile(args.labels) if args.labels is not None:
args.labels = findFile(args.labels)
# If config specified, try to load it as TensorFlow Object Detection API's pipeline. # If config specified, try to load it as TensorFlow Object Detection API's pipeline.
config = readTextMessage(args.config) config = readTextMessage(args.config)
@ -100,7 +98,10 @@ if args.labels:
labels = f.read().rstrip('\n').split('\n') labels = f.read().rstrip('\n').split('\n')
# Load a network # Load a network
net = cv.dnn.readNet(args.model, args.config, args.framework) engine = cv.dnn.ENGINE_AUTO
if args.backend != "default" or args.target != "cpu":
engine = cv.dnn.ENGINE_CLASSIC
net = cv.dnn.readNet(args.model, args.config, "", engine)
net.setPreferableBackend(get_backend_id(args.backend)) net.setPreferableBackend(get_backend_id(args.backend))
net.setPreferableTarget(get_target_id(args.target)) net.setPreferableTarget(get_target_id(args.target))
outNames = net.getUnconnectedOutLayersNames() outNames = net.getUnconnectedOutLayersNames()
@ -126,14 +127,10 @@ def postprocess(frame, outs):
frameHeight = frame.shape[0] frameHeight = frame.shape[0]
frameWidth = frame.shape[1] frameWidth = frame.shape[1]
layerNames = net.getLayerNames()
lastLayerId = net.getLayerId(layerNames[-1])
lastLayer = net.getLayer(lastLayerId)
classIds = [] classIds = []
confidences = [] confidences = []
boxes = [] boxes = []
if lastLayer.type == 'DetectionOutput': if args.postprocessing == 'ssd':
# Network produces output blob with a shape 1x1xNx7 where N is a number of # Network produces output blob with a shape 1x1xNx7 where N is a number of
# detections and an every detection is a vector of values # detections and an every detection is a vector of values
# [batchId, classId, confidence, left, top, right, bottom] # [batchId, classId, confidence, left, top, right, bottom]
@ -157,21 +154,12 @@ def postprocess(frame, outs):
classIds.append(int(detection[1]) - 1) # Skip background label classIds.append(int(detection[1]) - 1) # Skip background label
confidences.append(float(confidence)) confidences.append(float(confidence))
boxes.append([left, top, width, height]) boxes.append([left, top, width, height])
elif lastLayer.type == 'Region' or args.postprocessing == 'yolov8':
# Network produces output blob with a shape NxC where N is a number of elif args.postprocessing == 'darknet':
# detected objects and C is a number of classes + 4 where the first 4 box_scale_w = frameWidth
# numbers are [center_x, center_y, width, height] box_scale_h = frameHeight
if args.postprocessing == 'yolov8':
box_scale_w = frameWidth / args.width
box_scale_h = frameHeight / args.height
else:
box_scale_w = frameWidth
box_scale_h = frameHeight
for out in outs: for out in outs:
if args.postprocessing == 'yolov8':
out = out[0].transpose(1, 0)
for detection in out: for detection in out:
scores = detection[4:] scores = detection[4:]
if args.background_label_id >= 0: if args.background_label_id >= 0:
@ -188,13 +176,47 @@ def postprocess(frame, outs):
classIds.append(classId) classIds.append(classId)
confidences.append(float(confidence)) confidences.append(float(confidence))
boxes.append([left, top, width, height]) boxes.append([left, top, width, height])
elif args.postprocessing == 'yolov8' or args.postprocessing == 'yolov5':
# Network produces output blob with a shape NxC where N is a number of
# detected objects and C is a number of classes + 4 where the first 4
# numbers are [center_x, center_y, width, height]
box_scale_w = frameWidth / args.width
box_scale_h = frameHeight / args.height
for out in outs:
if args.postprocessing == 'yolov8':
out = out[0].transpose(1, 0)
else: # YOLOv5, no transposition needed
out = out[0]
for detection in out:
if args.postprocessing == 'yolov8':
scores = detection[4:]
obj_conf = 1
else:
scores = detection[5:]
obj_conf = detection[4]
classId = np.argmax(scores)
confidence = scores[classId]*obj_conf
if confidence > confThreshold:
center_x = int(detection[0] * box_scale_w)
center_y = int(detection[1] * box_scale_h)
width = int(detection[2] * box_scale_w)
height = int(detection[3] * box_scale_h)
left = int(center_x - width / 2)
top = int(center_y - height / 2)
classIds.append(classId)
confidences.append(float(confidence))
boxes.append([left, top, width, height])
else: else:
print('Unknown output layer type: ' + lastLayer.type) print('Unknown postprocessing method: ' + args.postprocessing)
exit() exit()
# NMS is used inside Region layer only on DNN_BACKEND_OPENCV for another backends we need NMS in sample # NMS is used inside Region layer only on DNN_BACKEND_OPENCV for another backends we need NMS in sample
# or NMS is required if number of outputs > 1 # or NMS is required if number of outputs > 1
if len(outNames) > 1 or (lastLayer.type == 'Region' or args.postprocessing == 'yolov8') and args.backend != cv.dnn.DNN_BACKEND_OPENCV: if len(outNames) > 1 or (args.postprocessing == 'darknet' or args.postprocessing == 'yolov8' or args.postprocessing == 'yolov5') and args.backend != cv.dnn.DNN_BACKEND_OPENCV:
indices = [] indices = []
classIds = np.array(classIds) classIds = np.array(classIds)
boxes = np.array(boxes) boxes = np.array(boxes)
@ -308,14 +330,11 @@ def processingThreadBody():
# Create a 4D blob from a frame. # Create a 4D blob from a frame.
inpWidth = args.width if args.width else frameWidth inpWidth = args.width if args.width else frameWidth
inpHeight = args.height if args.height else frameHeight inpHeight = args.height if args.height else frameHeight
blob = cv.dnn.blobFromImage(frame, size=(inpWidth, inpHeight), swapRB=args.rgb, ddepth=cv.CV_32F) blob = cv.dnn.blobFromImage(frame, scalefactor=args.scale, mean=args.mean, size=(inpWidth, inpHeight), swapRB=args.rgb, ddepth=cv.CV_32F)
processedFramesQueue.put(frame) processedFramesQueue.put(frame)
# Run a model # Run a model
net.setInput(blob, scalefactor=args.scale, mean=args.mean) net.setInput(blob)
if net.getLayer(0).outputNameToIndex('im_info') != -1: # Faster-RCNN or R-FCN
frame = cv.resize(frame, (inpWidth, inpHeight))
net.setInput(np.array([[inpHeight, inpWidth, 1.6]], dtype=np.float32), 'im_info')
if asyncN: if asyncN:
futureOutputs.append(net.forwardAsync()) futureOutputs.append(net.forwardAsync())
@ -385,9 +404,9 @@ else:
inpWidth = args.width if args.width else frameWidth inpWidth = args.width if args.width else frameWidth
inpHeight = args.height if args.height else frameHeight inpHeight = args.height if args.height else frameHeight
blob = cv.dnn.blobFromImage(frame, size=(inpWidth, inpHeight), swapRB=args.rgb, ddepth=cv.CV_32F) blob = cv.dnn.blobFromImage(frame, scalefactor=args.scale, mean=args.mean, size=(inpWidth, inpHeight), swapRB=args.rgb, ddepth=cv.CV_32F)
net.setInput(blob, scalefactor=args.scale, mean=args.mean) net.setInput(blob)
outs = net.forward(outNames) outs = net.forward(outNames)
boxes, classIds, confidences, indices = postprocess(frame, outs) boxes, classIds, confidences, indices = postprocess(frame, outs)