mirror of
https://github.com/deepinsight/insightface.git
synced 2026-07-13 07:47:49 +00:00
Update InspireFace to 1.0.1
This commit is contained in:
JingyuYan
@@ -8,178 +8,4 @@
|
||||
#include "../test_helper/test_help.h"
|
||||
#include "feature_hub/feature_hub.h"
|
||||
|
||||
using namespace inspire;
|
||||
|
||||
TEST_CASE("test_FaceFeatureManagement", "[face_feature]") {
|
||||
DRAW_SPLIT_LINE
|
||||
TEST_PRINT_OUTPUT(true);
|
||||
|
||||
SECTION("FeatureCURD") {
|
||||
DRAW_SPLIT_LINE
|
||||
// Initialize
|
||||
FaceContext ctx;
|
||||
CustomPipelineParameter param;
|
||||
param.enable_recognition = true;
|
||||
auto ret = ctx.Configuration(DetectMode::DETECT_MODE_IMAGE, 1, param);
|
||||
REQUIRE(ret == HSUCCEED);
|
||||
|
||||
FEATURE_HUB->PrintFeatureMatrixInfo();
|
||||
|
||||
// Know the location of 'kunkun' in advance
|
||||
int32_t KunkunIndex = 795;
|
||||
// Prepare a face photo in advance and extract the features
|
||||
auto image = cv::imread(GET_DATA("images/kun.jpg"));
|
||||
CameraStream stream;
|
||||
stream.SetDataFormat(BGR);
|
||||
stream.SetRotationMode(ROTATION_0);
|
||||
stream.SetDataBuffer(image.data, image.rows, image.cols);
|
||||
ret = ctx.FaceDetectAndTrack(stream);
|
||||
REQUIRE(ret == HSUCCEED);
|
||||
// Face detection
|
||||
ctx.FaceDetectAndTrack(stream);
|
||||
const auto &faces = ctx.GetTrackingFaceList();
|
||||
REQUIRE(faces.size() > 0);
|
||||
// Feature extraction of "Kunkun" was carried out
|
||||
Embedded feature;
|
||||
ret = ctx.FaceRecognitionModule()->FaceExtract(stream, faces[0], feature);
|
||||
CHECK(ret == HSUCCEED);
|
||||
|
||||
// Import face feature vectors in batches
|
||||
String mat_path = GET_DATA("test_faceset/test_faces_A1.npy");
|
||||
String tags_path = GET_DATA("test_faceset/test_faces_A1.txt");
|
||||
auto result = LoadMatrixAndTags(mat_path, tags_path);
|
||||
// Gets the feature matrix and label names
|
||||
EmbeddedList featureMatrix = result.first;
|
||||
std::vector<std::string> tagNames = result.second;
|
||||
REQUIRE(featureMatrix.size() == 3000);
|
||||
REQUIRE(tagNames.size() == 3000);
|
||||
REQUIRE(featureMatrix[0].size() == 512);
|
||||
|
||||
for (int i = 0; i < featureMatrix.size(); ++i) {
|
||||
auto &feat = featureMatrix[i];
|
||||
auto ret = FEATURE_HUB->RegisterFaceFeature(feat, i, tagNames[i], i);
|
||||
CHECK(ret == HSUCCEED);
|
||||
}
|
||||
|
||||
std::cout << std::endl;
|
||||
REQUIRE(FEATURE_HUB->GetFaceFeatureCount() == 3000);
|
||||
spdlog::trace("All 3000 Faces embedded vector are loaded");
|
||||
|
||||
// Prepare a face photo to search through the library
|
||||
SearchResult searchResult;
|
||||
ret = FEATURE_HUB->SearchFaceFeature(feature, searchResult, 0.5f);
|
||||
REQUIRE(ret == HSUCCEED);
|
||||
CHECK(searchResult.index != -1);
|
||||
CHECK(searchResult.index == KunkunIndex);
|
||||
CHECK(searchResult.tag == "Kunkun");
|
||||
CHECK(searchResult.score == Approx(0.76096).epsilon(1e-3));
|
||||
spdlog::info("Find Kunkun -> Location ID: {}, Confidence: {}, Tag: {}", searchResult.index, searchResult.score, searchResult.tag.c_str());
|
||||
// Save "Kunkun"'s library features and so on
|
||||
Embedded KunkunFeature;
|
||||
ret = FEATURE_HUB->GetFaceFeature(KunkunIndex, KunkunFeature);
|
||||
REQUIRE(ret == HSUCCEED);
|
||||
|
||||
// The features of "Kunkun" library corresponding to those found above are deleted from the face library
|
||||
ret = FEATURE_HUB->DeleteFaceFeature(searchResult.index);
|
||||
CHECK(ret == HSUCCEED);
|
||||
// In search once
|
||||
SearchResult secondSearchResult;
|
||||
ret = FEATURE_HUB->SearchFaceFeature(feature, secondSearchResult, 0.5f);
|
||||
REQUIRE(ret == HSUCCEED);
|
||||
CHECK(secondSearchResult.index == -1);
|
||||
spdlog::info("Kunkun被删除了无法找到: {}, {}", secondSearchResult.index, secondSearchResult.tag);
|
||||
|
||||
// Just take a random place and change the eigenvector for that place and put "Kunkun" back in there
|
||||
auto newIndex = 2888;
|
||||
// Try inserting an unused location first
|
||||
ret = FEATURE_HUB->UpdateFaceFeature(KunkunFeature, 3001, "Chicken", 3001);
|
||||
REQUIRE(ret == HERR_SESS_REC_BLOCK_UPDATE_FAILURE);
|
||||
ret = FEATURE_HUB->UpdateFaceFeature(KunkunFeature, newIndex, "Chicken", 3001);
|
||||
REQUIRE(ret == HSUCCEED);
|
||||
SearchResult thirdlySearchResult;
|
||||
ret = FEATURE_HUB->SearchFaceFeature(feature, thirdlySearchResult, 0.5f);
|
||||
REQUIRE(ret == HSUCCEED);
|
||||
CHECK(thirdlySearchResult.index != -1);
|
||||
CHECK(thirdlySearchResult.index == newIndex);
|
||||
CHECK(thirdlySearchResult.tag == "Chicken");
|
||||
spdlog::info("Find Kunkun again -> New Location ID: {}, Confidence: {}, Tag: {}", thirdlySearchResult.index, thirdlySearchResult.score, thirdlySearchResult.tag.c_str());
|
||||
|
||||
}
|
||||
|
||||
#if ENABLE_BENCHMARK
|
||||
SECTION("FeatureSearchBenchmark") {
|
||||
DRAW_SPLIT_LINE
|
||||
|
||||
// Initialize
|
||||
FaceContext ctx;
|
||||
CustomPipelineParameter param;
|
||||
param.enable_recognition = true;
|
||||
auto ret = ctx.Configuration(DetectMode::DETECT_MODE_IMAGE, 1, param);
|
||||
REQUIRE(ret == HSUCCEED);
|
||||
|
||||
FEATURE_HUB->PrintFeatureMatrixInfo();
|
||||
|
||||
// Import face feature vectors in batches
|
||||
String mat_path = GET_DATA("test_faceset/test_faces_A1.npy");
|
||||
String tags_path = GET_DATA("test_faceset/test_faces_A1.txt");
|
||||
auto result = LoadMatrixAndTags(mat_path, tags_path);
|
||||
// Gets the feature matrix and label names
|
||||
EmbeddedList featureMatrix = result.first;
|
||||
std::vector<std::string> tagNames = result.second;
|
||||
REQUIRE(featureMatrix.size() == 3000);
|
||||
REQUIRE(tagNames.size() == 3000);
|
||||
REQUIRE(featureMatrix[0].size() == 512);
|
||||
|
||||
for (int i = 0; i < featureMatrix.size(); ++i) {
|
||||
auto &feat = featureMatrix[i];
|
||||
auto ret = FEATURE_HUB->RegisterFaceFeature(feat, i, tagNames[i], i);
|
||||
CHECK(ret == HSUCCEED);
|
||||
}
|
||||
|
||||
std::cout << std::endl;
|
||||
REQUIRE(FEATURE_HUB->GetFaceFeatureCount() == 3000);
|
||||
spdlog::trace("3000个特征向量全部载入");
|
||||
|
||||
// Prepare a picture of a face
|
||||
auto image = cv::imread(GET_DATA("images/face_sample.png"));
|
||||
CameraStream stream;
|
||||
stream.SetDataFormat(BGR);
|
||||
stream.SetRotationMode(ROTATION_0);
|
||||
stream.SetDataBuffer(image.data, image.rows, image.cols);
|
||||
ret = ctx.FaceDetectAndTrack(stream);
|
||||
REQUIRE(ret == HSUCCEED);
|
||||
// Face detection
|
||||
ctx.FaceDetectAndTrack(stream);
|
||||
const auto &faces = ctx.GetTrackingFaceList();
|
||||
REQUIRE(faces.size() > 0);
|
||||
// Feature extraction of "kunkun" was carried out
|
||||
Embedded feature;
|
||||
ret = ctx.FaceRecognitionModule()->FaceExtract(stream, faces[0], feature);
|
||||
CHECK(ret == HSUCCEED);
|
||||
|
||||
// Insert the face further back
|
||||
auto regIndex = 4000;
|
||||
ret = FEATURE_HUB->RegisterFaceFeature(feature, regIndex, "test", 4000);
|
||||
REQUIRE(ret == HSUCCEED);
|
||||
|
||||
const auto loop = 1000;
|
||||
double total = 0.0f;
|
||||
spdlog::info("Start performing {} searches: ", loop);
|
||||
auto out = (double) cv::getTickCount();
|
||||
for (int i = 0; i < loop; ++i) {
|
||||
|
||||
// Prepare a face photo to look it up from the library
|
||||
SearchResult searchResult;
|
||||
auto timeStart = (double) cv::getTickCount();
|
||||
ret = FEATURE_HUB->SearchFaceFeature(feature, searchResult, 0.5f);
|
||||
double cost = ((double) cv::getTickCount() - timeStart) / cv::getTickFrequency() * 1000;
|
||||
REQUIRE(ret == HSUCCEED);
|
||||
CHECK(searchResult.index == regIndex);
|
||||
total += cost;
|
||||
}
|
||||
auto end = ((double) cv::getTickCount() - out) / cv::getTickFrequency() * 1000;
|
||||
|
||||
spdlog::info("Execute {} times Total Cost: {}ms, Average Cost: {}ms", loop, end, total / loop);
|
||||
}
|
||||
#endif
|
||||
}
|
||||
using namespace inspire;
|
||||
Reference in New Issue
Block a user