计算速度优化，摄像头读取逻辑改变，分类器更新。

2019-07-28 15:59:01 +08:00
parent a5259cbd1f
commit 5a0fdb30af
24 changed files with 17667 additions and 115529 deletions
--- a/.gitignore
+++ b/.gitignore
@@ -4,6 +4,7 @@ build
 Mark
 armor_box_photo
 tools/TrainCNN/.idea
+tools/TrainCNN/model
 tools/TrainCNN/__pycache__
 others/include/config/config.h
 .DS_Store
--- a/armor/include/armor_finder/armor_finder.h
+++ b/armor/include/armor_finder/armor_finder.h
@@ -51,10 +51,11 @@ extern std::map<string, int> prior_red;
 class LightBlob {
 public:
    cv::RotatedRect rect;   //灯条位置
+    double areaRatio;
    double length;          //灯条长度
    uint8_t blob_color;      //灯条颜色

-    LightBlob(cv::RotatedRect &r, uint8_t color) : rect(r), blob_color(color) {
+    LightBlob(cv::RotatedRect &r, double ratio, uint8_t color) : rect(r), areaRatio(ratio), blob_color(color) {
        length = max(rect.size.height, rect.size.width);
    };
    LightBlob() = default;
--- a/armor/include/armor_finder/classifier/classifier.h
+++ b/armor/include/armor_finder/classifier/classifier.h
@@ -18,8 +18,8 @@ class Classifier {
 private:
    bool state;

-    vector<vector<MatrixXd>> conv1_w, conv2_w;
-    vector<double> conv1_b, conv2_b;
+    vector<vector<MatrixXd>> conv1_w, conv2_w, conv3_w;
+    vector<double> conv1_b, conv2_b, conv3_b;
    MatrixXd fc1_w, fc2_w;
    VectorXd fc1_b, fc2_b;

@@ -30,8 +30,10 @@ private:

    MatrixXd softmax(const MatrixXd &input);
    MatrixXd relu(const MatrixXd &input);
+    MatrixXd leaky_relu(const MatrixXd &input, float alpha);
    vector<vector<MatrixXd>> apply_bias(const vector<vector<MatrixXd>> &input, const vector<double> &bias);
    vector<vector<MatrixXd>> relu(const vector<vector<MatrixXd>> &input);
+    vector<vector<MatrixXd>> leaky_relu(const vector<vector<MatrixXd>> &input, float alpha);
    vector<vector<MatrixXd>> max_pool(const vector<vector<MatrixXd>> &input, int size);
    vector<vector<MatrixXd>> mean_pool(const vector<vector<MatrixXd>> &input, int size);
    vector<vector<MatrixXd>> pand(const vector<vector<MatrixXd>> &input, int val);
--- a/armor/src/armor_finder/anti_top/anti_top.cpp
+++ b/armor/src/armor_finder/anti_top/anti_top.cpp
@@ -3,12 +3,9 @@
 //

 #include <armor_finder/armor_finder.h>
+#include <additions/additions.h>
 #include <log.h>

-static double getTimeIntervalms(const timeval& now, const timeval &last){
-    return (now.tv_sec-last.tv_sec)*1000.0 + (now.tv_usec-last.tv_usec)/1000.0;
-}
-
 void ArmorFinder::antiTop() {
    static double top_periodms = 0;
    static double last_top_periodms = 0;
--- a/armor/src/armor_finder/armor_finder.cpp
+++ b/armor/src/armor_finder/armor_finder.cpp
@@ -58,8 +58,8 @@ ArmorFinder::ArmorFinder(uint8_t &color, Serial &u, const string &paras_folder,
 }

 void ArmorFinder::run(cv::Mat &src) {
-//    stateSearchingTarget(src);                    // for debug
-//    goto end;
+    stateSearchingTarget(src);                    // for debug
+    goto end;
    switch (state) {
        case SEARCHING_STATE:
            if (stateSearchingTarget(src)) {
--- a/armor/src/armor_finder/classifier/classifier.cpp
+++ b/armor/src/armor_finder/classifier/classifier.cpp
@@ -162,6 +162,12 @@ MatrixXd Classifier::relu(const MatrixXd &input){
    });
 }

+MatrixXd Classifier::leaky_relu(const MatrixXd &input, float alpha){
+    return input.unaryExpr([&](double val){
+        return (val>0)?(val):(alpha*val);
+    });
+}
+
 vector<vector<MatrixXd>> Classifier::relu(const vector<vector<MatrixXd>> &input){
    vector<vector<MatrixXd>> result;
    for(int samples=0; samples<input.size(); samples++){
@@ -174,6 +180,18 @@ vector<vector<MatrixXd>> Classifier::relu(const vector<vector<MatrixXd>> &input)
    return result;
 }

+vector<vector<MatrixXd>> Classifier::leaky_relu(const vector<vector<MatrixXd>> &input, float alpha){
+    vector<vector<MatrixXd>> result;
+    for(int samples=0; samples<input.size(); samples++){
+        vector<MatrixXd> sub;
+        for(int channels=0; channels<input[0].size(); channels++){
+            sub.emplace_back(leaky_relu(input[samples][channels], alpha));
+        }
+        result.emplace_back(sub);
+    }
+    return result;
+}
+
 vector<vector<MatrixXd>> Classifier::pand(const vector<vector<MatrixXd>> &input, int val){
    vector<vector<MatrixXd>> result;
    for(int sample=0; sample<input.size(); sample++){
@@ -254,6 +272,8 @@ Classifier::Classifier(const string &folder) : state(true){
    conv1_b = load_conv_b(folder+"conv1_b");
    conv2_w = load_conv_w(folder+"conv2_w");
    conv2_b = load_conv_b(folder+"conv2_b");
+    conv3_w = load_conv_w(folder+"conv3_w");
+    conv3_b = load_conv_b(folder+"conv3_b");
    fc1_w = load_fc_w(folder+"fc1_w");
    fc1_b = load_fc_b(folder+"fc1_b");
    fc2_w = load_fc_w(folder+"fc2_w");
@@ -273,7 +293,8 @@ MatrixXd Classifier::calculate(const vector<vector<MatrixXd>> &input) {
    vector<vector<MatrixXd>> pool1_result = mean_pool(conv1_result, 2);
    vector<vector<MatrixXd>> conv2_result = relu(apply_bias(conv2(conv2_w, pool1_result), conv2_b));
    vector<vector<MatrixXd>> pool2_result = mean_pool(conv2_result, 2);
-    MatrixXd flattened = flatten(pool2_result);
+    vector<vector<MatrixXd>> conv3_result = relu(apply_bias(conv2(conv3_w, pool2_result), conv3_b));
+    MatrixXd flattened = flatten(conv3_result);
    MatrixXd y1 = fc1_w * flattened;
    y1.colwise() += fc1_b;
    MatrixXd fc1 = relu(y1);
@@ -303,7 +324,7 @@ int Classifier::operator()(const cv::Mat &image) {
 //    cout << result << "==============" <<endl;
    MatrixXd::Index minRow, minCol;
    result.maxCoeff(&minRow, &minCol);
-    if(result(minRow, minCol) > 0.9){
+    if(result(minRow, minCol) > 0.90){
        return minRow;
    }else{
        return 0;
--- a/armor/src/armor_finder/find/find_armor_box.cpp
+++ b/armor/src/armor_finder/find/find_armor_box.cpp
@@ -6,6 +6,7 @@
 #include <show_images/show_images.h>
 #include <options/options.h>
 #include <opencv2/highgui.hpp>
+#define DO_NOT_CNT_TIME
 #include <log.h>


@@ -27,7 +28,7 @@ static bool lengthJudge(const LightBlob &light_blob_i, const LightBlob &light_bl
    double side_length;
    cv::Point2f centers = light_blob_i.rect.center - light_blob_j.rect.center;
    side_length = sqrt(centers.ddot(centers));
-    return (side_length / light_blob_i.length < 8 && side_length / light_blob_i.length > 0.5);
+    return (side_length / light_blob_i.length < 10 && side_length / light_blob_i.length > 0.5);
 }

 static bool lengthRatioJudge(const LightBlob &light_blob_i, const LightBlob &light_blob_j) {
@@ -99,6 +100,7 @@ bool matchArmorBoxes(const cv::Mat &src, const LightBlobs &light_blobs, ArmorBox
            if (min_x < 0 || max_x > src.cols || min_y < 0 || max_y > src.rows) {
                continue;
            }
+            if((max_x-min_x)/(max_y-min_y) < 0.8) continue;
            LightBlobs pair_blobs = {light_blobs.at(i), light_blobs.at(j)};
            armor_boxes.emplace_back(
                    cv::Rect2d(min_x, min_y, max_x - min_x, max_y - min_y),
@@ -116,31 +118,35 @@ bool ArmorFinder::findArmorBox(const cv::Mat &src, ArmorBox &box) {

    box.rect = cv::Rect2d(0, 0, 0, 0);
    box.id = -1;
-
+    CNT_TIME("blob", {
        if (!findLightBlobs(src, light_blobs)) {
            return false;
        }
+    });
    if (show_light_blobs && src.size() == cv::Size(640, 480)) {
        showLightBlobs("light_blobs", src, light_blobs);
        cv::waitKey(1);
    }
-
+    CNT_TIME("boxes",{
        if (!matchArmorBoxes(src, light_blobs, armor_boxes, enemy_color)) {
    //        cout << "Box fail!" << endl;
            return false;
        }
+    });
    if (show_armor_boxes && src.size() == cv::Size(640, 480)) {
        showArmorBoxes("boxes", src, armor_boxes);
        cv::waitKey(1);
    }

    if (classifier && use_classifier) {
+        CNT_TIME("classify: %d", {
            for (auto &armor_box : armor_boxes) {
                cv::Mat roi = src(armor_box.rect).clone();
                cv::resize(roi, roi, cv::Size(48, 36));
                int c = classifier(roi);
                armor_box.id = c;
            }
+        }, armor_boxes.size());
        sort(armor_boxes.begin(), armor_boxes.end());
        if(armor_boxes[0].id != 0){
            box = armor_boxes[0];
--- a/armor/src/armor_finder/find/find_light_blobs.cpp
+++ b/armor/src/armor_finder/find/find_light_blobs.cpp
@@ -95,6 +95,11 @@ static double get_blob_color_opt(const cv::Mat &src, cv::RotatedRect blobPos) {
    }
 }

+static bool isSameBlob(LightBlob blob1, LightBlob blob2) {
+    auto dist = blob1.rect.center - blob2.rect.center;
+    return (dist.x * dist.x + dist.y * dist.y) < 9;
+}
+
 static void imagePreProcess(cv::Mat &src) {
    static cv::Mat kernel_erode = getStructuringElement(cv::MORPH_RECT, cv::Size(3, 5));
    erode(src, src, kernel_erode);
@@ -126,7 +131,6 @@ bool ArmorFinder::findLightBlobs(const cv::Mat &src, LightBlobs &light_blobs) {
    }


-
    cv::threshold(color_channel, src_bin_light, 200, 255, CV_THRESH_BINARY); // 二值化对应通道
    if (src_bin_light.empty()) return false;
    imagePreProcess(src_bin_light);                                  // 开闭运算
@@ -141,19 +145,55 @@ bool ArmorFinder::findLightBlobs(const cv::Mat &src, LightBlobs &light_blobs) {
    }

    std::vector<std::vector<cv::Point>> light_contours_light, light_contours_dim;
-    cv::findContours(src_bin_light, light_contours_light, CV_RETR_LIST, CV_CHAIN_APPROX_NONE);
-    cv::findContours(src_bin_dim, light_contours_dim, CV_RETR_LIST, CV_CHAIN_APPROX_NONE);
-    for (auto &light_contour : light_contours_light) {
-        cv::RotatedRect rect = cv::minAreaRect(light_contour);
-        if (isValidLightBlob(light_contour, rect)) {
-            light_blobs.emplace_back(rect, get_blob_color(src, rect));
+    LightBlobs light_blobs_light, light_blobs_dim;
+    std::vector<cv::Vec4i> hierarchy_light, hierarchy_dim;
+    cv::findContours(src_bin_light, light_contours_light, hierarchy_light, CV_RETR_CCOMP, CV_CHAIN_APPROX_NONE);
+    cv::findContours(src_bin_dim, light_contours_dim, hierarchy_dim, CV_RETR_CCOMP, CV_CHAIN_APPROX_NONE);
+    for (int i = 0; i < light_contours_light.size(); i++) {
+        if (hierarchy_light[i][2] == -1) {
+            cv::RotatedRect rect = cv::minAreaRect(light_contours_light[i]);
+            if (isValidLightBlob(light_contours_light[i], rect)) {
+                light_blobs_light.emplace_back(
+                        rect, areaRatio(light_contours_light[i], rect), get_blob_color(src, rect)
+                );
            }
        }
-    for (auto &light_contour : light_contours_dim) {
-        cv::RotatedRect rect = cv::minAreaRect(light_contour);
-        if (isValidLightBlob(light_contour, rect)) {
-            light_blobs.emplace_back(rect, get_blob_color(src, rect));
    }
+    for (int i = 0; i < light_contours_dim.size(); i++) {
+        if (hierarchy_dim[i][2] == -1) {
+            cv::RotatedRect rect = cv::minAreaRect(light_contours_dim[i]);
+            if (isValidLightBlob(light_contours_dim[i], rect)) {
+                light_blobs_dim.emplace_back(
+                        rect, areaRatio(light_contours_dim[i], rect), get_blob_color(src, rect)
+                );
+            }
+        }
+    }
+    vector<int> light_to_remove, dim_to_remove;
+    for (int l = 0; l != light_blobs_light.size(); l++) {
+        for (int d = 0; d != light_blobs_dim.size(); d++) {
+            if (isSameBlob(light_blobs_light[l], light_blobs_dim[d])) {
+                if (light_blobs_light[l].areaRatio > light_blobs_dim[d].areaRatio) {
+                    dim_to_remove.emplace_back(d);
+                } else {
+                    light_to_remove.emplace_back(l);
+                }
+            }
+        }
+    }
+    sort(light_to_remove.begin(), light_to_remove.end(), [](int a, int b) { return a > b; });
+    sort(dim_to_remove.begin(), dim_to_remove.end(), [](int a, int b) { return a > b; });
+    for (auto x : light_to_remove) {
+        light_blobs_light.erase(light_blobs_light.begin() + x);
+    }
+    for (auto x : dim_to_remove) {
+        light_blobs_dim.erase(light_blobs_dim.begin() + x);
+    }
+    for (const auto &light : light_blobs_light) {
+        light_blobs.emplace_back(light);
+    }
+    for (const auto &dim : light_blobs_dim) {
+        light_blobs.emplace_back(dim);
    }
    return light_blobs.size() >= 2;
 }
--- a/main.cpp
+++ b/main.cpp
@@ -20,7 +20,7 @@
 #include <additions/additions.h>
 #include <config/setconfig.h>

-#define DO_NOT_CNT_TIME
+//#define DO_NOT_CNT_TIME

 #include <log.h>

@@ -63,8 +63,8 @@ int main(int argc, char *argv[]) {
    while (true) {
        // 打开视频源
        if (from_camera) {
-            video_gimbal = new CameraWrapper(ARMOR_CAMERA_GAIN, 0/*, "armor"*/);
-            video_chassis = new CameraWrapper(ENERGY_CAMERA_GAIN, 0/*, "energy"*/);
+            video_gimbal = new CameraWrapper(ARMOR_CAMERA_GAIN, 2/*, "armor"*/);
+            video_chassis = new CameraWrapper(ENERGY_CAMERA_GAIN, 2/*, "energy"*/);
        } else {
            video_gimbal = new VideoWrapper("/home/sun/项目/energy_video/7.27.avi");
            video_chassis = new VideoWrapper("/home/sun/项目/energy_video/7.27.avi");
@@ -93,14 +93,14 @@ int main(int argc, char *argv[]) {
        bool ok = true;
        cout << "start running" << endl;
        do {
-//            CNT_TIME("Total", {
+            CNT_TIME("Total", {
            if (mcuData.state == BIG_ENERGY_STATE) {//大能量机关模式
                if (last_state != BIG_ENERGY_STATE) {//若上一帧不是大能量机关模式，即刚往完成切换，则需要初始化
                    LOGM(STR_CTR(WORD_BLUE, "Start Big Energy!"));
                    destroyAllWindows();
                    if (from_camera) {
                        delete video_gimbal;
-                        video_gimbal = new CameraWrapper(ENERGY_CAMERA_GAIN, 0/*, "armor"*/);
+                        video_gimbal = new CameraWrapper(ENERGY_CAMERA_GAIN, 2/*, "armor"*/);
                        if (video_gimbal->init()) {
                            LOGM("video_gimbal source initialization successfully.");
                        } else {
@@ -130,7 +130,7 @@ int main(int argc, char *argv[]) {
                    destroyAllWindows();
                    if (from_camera) {
                        delete video_gimbal;
-                        video_gimbal = new CameraWrapper(ENERGY_CAMERA_GAIN, 0/*, "armor"*/);
+                        video_gimbal = new CameraWrapper(ENERGY_CAMERA_GAIN, 2/*, "armor"*/);
                        if (video_gimbal->init()) {
                            LOGM("video_gimbal source initialization successfully.");
                        } else {
@@ -154,7 +154,7 @@ int main(int argc, char *argv[]) {
                    destroyAllWindows();
                    if (from_camera) {
                        delete video_gimbal;
-                        video_gimbal = new CameraWrapper(ARMOR_CAMERA_GAIN, 0/*, "armor"*/);
+                        video_gimbal = new CameraWrapper(ARMOR_CAMERA_GAIN, 2/*, "armor"*/);
                        if (video_gimbal->init()) {
                            LOGM("video_gimbal source initialization successfully.");
                        } else {
@@ -163,19 +163,23 @@ int main(int argc, char *argv[]) {
                    }
                }
                last_state = mcuData.state;
-                ok = checkReconnect(video_gimbal->read(gimbal_src));
+                CNT_TIME(STR_CTR(WORD_GREEN, "read img"), {
+                    if(!checkReconnect(video_gimbal->read(gimbal_src))) continue;
+                });
 #ifdef GIMBAL_FLIP_MODE
                flip(gimbal_src, gimbal_src, GIMBAL_FLIP_MODE);
 #endif
+//                CNT_TIME("something whatever", {
                    if (!from_camera) extract(gimbal_src);
                    if (save_video) saveVideos(gimbal_src);
                    if (show_origin) showOrigin(gimbal_src);
-                CNT_TIME("Armor Time", {
+//                });
+                CNT_TIME(STR_CTR(WORD_CYAN, "Armor Time"), {
                    armorFinder.run(gimbal_src);
                });
            }
 //                cv::waitKey(0);
-//            });
+            });
        } while (ok);
        delete video_gimbal;
        video_gimbal = nullptr;
--- a/others/include/additions/additions.h
+++ b/others/include/additions/additions.h
@@ -6,6 +6,7 @@
 #define _ADDITIONS_H_

 #include <stdint.h>
+#include <sys/time.h>
 #include <serial/serial.h>

 struct mcu_data{
@@ -30,5 +31,6 @@ void showOrigin(const cv::Mat &gimbal_src, const cv::Mat &chassis_src);
 void showOrigin(const cv::Mat &gimbal_src);
 void extract(cv::Mat &gimbal_src, cv::Mat &chassis_src);
 void extract(cv::Mat &gimbal_src);
+double getTimeIntervalms(const timeval& now, const timeval &last);

 #endif /* _ADDITIONS_H_ */
--- a/others/include/camera/camera_wrapper.h
+++ b/others/include/camera/camera_wrapper.h
@@ -9,11 +9,12 @@

 #include <stdio.h>
 #include <iostream>
+#include <thread>
 #include "opencv2/core/core.hpp"
 #include "opencv2/highgui/highgui.hpp"
 #include <opencv2/imgproc/imgproc.hpp>
-
 #include "camera/wrapper_head.h"
+
 #ifdef Windows
    #include "camera/CameraApi.h"
 #elif defined(Linux) || defined(Darwin)
@@ -21,10 +22,13 @@
 #endif

 class CameraWrapper: public WrapperHead {
+    friend void cameraCallback(CameraHandle hCamera, BYTE *pFrameBuffer, tSdkFrameHead* pFrameHead,PVOID pContext);
 private:
    const std::string name;
    int mode;

+    bool init_done;
+
    unsigned char* rgb_buffer;
    int camera_cnts;
    int camera_status;
@@ -38,6 +42,11 @@ private:
    IplImage* iplImage;
    int channel;

+    cv::Mat src_queue[2];
+    volatile int qhead;
+    volatile int qtail;
+
+    std::thread *readThread;
 public:
    int gain;

@@ -48,6 +57,8 @@ public:
    bool read(cv::Mat& src) final;
    bool readRaw(cv::Mat& src);
    bool readProcessed(cv::Mat& src);
+    bool readCallback(cv::Mat& src);
+    bool readWithThread(cv::Mat &src);
    bool changeBrightness(int brightness);
 //    bool once
 };
--- a/others/include/log.h
+++ b/others/include/log.h
@@ -146,7 +146,7 @@
 			gettimeofday(&ts, NULL); \
 			codes; \
 			gettimeofday(&te, NULL); \
-			LOGM(tag": %.1lfms", (te.tv_sec-ts.tv_sec)*1000.0+(te.tv_usec-ts.tv_usec)/1000.0); \
+			LOGM(tag": %.1lfms", ##__VA_ARGS__, (te.tv_sec-ts.tv_sec)*1000.0+(te.tv_usec-ts.tv_usec)/1000.0); \
 		}while (0)
    #else
        #warning "Unsupport plantform for CNT_TIME"
--- a/others/src/additions/additions.cpp
+++ b/others/src/additions/additions.cpp
@@ -169,3 +169,7 @@ void extract(cv::Mat &gimbal_src) {//图像预处理，将视频切成640×480
        gimbal_src = gimbal_src(Rect(0, (width - 480) / 2, 640, 480));
    }
 }
+
+double getTimeIntervalms(const timeval& now, const timeval &last){
+    return (now.tv_sec-last.tv_sec)*1000.0 + (now.tv_usec-last.tv_usec)/1000.0;
+}
--- a/others/src/camera/camera_wrapper.cpp
+++ b/others/src/camera/camera_wrapper.cpp
@@ -4,6 +4,7 @@

 #include <camera/camera_wrapper.h>
 #include <log.h>
+#include <additions/additions.h>
 #include <options/options.h>
 #include <config/setconfig.h>

@@ -15,15 +16,33 @@ using namespace cv;

 CameraWrapper::CameraWrapper(int gain, int camera_mode, const std::string &n) :
        name(n),
+        init_done(false),
        mode(camera_mode),
        camera_cnts(2),
        camera_status(-1),
        iplImage(nullptr),
        rgb_buffer(nullptr),
        channel(3),
-        gain(gain){
+        gain(gain),
+        qhead(0),
+        qtail(0),
+        readThread(nullptr){
 }

+void cameraCallback(CameraHandle hCamera, BYTE *pFrameBuffer, tSdkFrameHead* pFrameHead,PVOID pContext){
+    CameraWrapper *c = (CameraWrapper*)pContext;
+    CameraImageProcess(hCamera, pFrameBuffer, c->rgb_buffer, pFrameHead);
+    auto iplImage = cvCreateImageHeader(cvSize(pFrameHead->iWidth, pFrameHead->iHeight), IPL_DEPTH_8U, c->channel);
+    cvSetData(iplImage, c->rgb_buffer, pFrameHead->iWidth * c->channel);  //此处只是设置指针，无图像块数据拷贝，不需担心转换效率
+    c->src_queue[c->qhead] = cv::cvarrToMat(iplImage).clone();
+    if((c->qhead+1)%2 == c->qtail){
+        c->qhead = (c->qhead+1)%2;
+        c->qtail = (c->qtail+1)%2;
+    } else {
+        c->qhead = (c->qhead+1)%2;
+    }
+//    LOGM("Get image, [%d %d]", c->qhead, c->qtail);
+}

 bool CameraWrapper::init() {
    CameraSdkInit(1);
@@ -76,6 +95,8 @@ bool CameraWrapper::init() {
    }
    LOGM("successfully loaded %s!", filepath);
 #elif defined(Linux)
+    CameraReadParameterFromFile(h_camera, PROJECT_DIR"/others/MV-UB31-Group0.config");
+    CameraLoadParameter(h_camera, PARAMETER_TEAM_A);
    CameraSetAeState(h_camera, false);
    CameraSetExposureTime(h_camera, CAMERA_EXPOSURE * 1000);
 #ifndef WITH_TIME_BASED_CAMERA_GAIN
@@ -106,14 +127,13 @@ bool CameraWrapper::init() {
    }
    CameraSetAnalogGain(h_camera, gain);
 #endif
-    if (mode == 0) {
-        CameraSetGain(h_camera, CAMERA_RED_GAIN, CAMERA_GREEN_GAIN, CAMERA_BLUE_GAIN);
-        CameraSetLutMode(h_camera, LUTMODE_PRESET);
-    }
+
 #endif
    double t;
+    int g;
    CameraGetExposureTime(h_camera, &t);
-    LOGM("Exposure time: %lfms", t / 1000.0);
+    CameraGetAnalogGain(h_camera, &g);
+    LOGM("Exposure time: %lfms, gain:%d", t / 1000.0, g);
    /*让SDK进入工作模式，开始接收来自相机发送的图像
    数据。如果当前相机是触发模式，则需要接收到
    触发帧以后才会更新图像。    */
@@ -135,6 +155,10 @@ bool CameraWrapper::init() {
        CameraSetIspOutFormat(h_camera, CAMERA_MEDIA_TYPE_BGR8);
        LOGM("camera %s color ", camera_name);
    }
+    if(mode == 2){
+        CameraSetCallbackFunction(h_camera, cameraCallback, this, nullptr);
+    }
+    init_done = true;
    return true;
 }

@@ -144,12 +168,18 @@ bool CameraWrapper::changeBrightness(int brightness) {
 }

 bool CameraWrapper::read(cv::Mat &src) {
+    if(init_done) {
        if (mode == 0)return readProcessed(src);
        if (mode == 1)return readRaw(src);
+        if (mode == 2)return readCallback(src);
+        if (mode == 3)return readWithThread(src);
+    } else {
+        return false;
+    }
 }

 bool CameraWrapper::readRaw(cv::Mat &src) {
-    if (CameraGetImageBuffer(h_camera, &frame_info, &pby_buffer, 1000) == CAMERA_STATUS_SUCCESS) {
+    if (CameraGetImageBuffer(h_camera, &frame_info, &pby_buffer, 500) == CAMERA_STATUS_SUCCESS) {
        if (iplImage) {
            cvReleaseImageHeader(&iplImage);
        }
@@ -173,7 +203,7 @@ bool CameraWrapper::readRaw(cv::Mat &src) {

 bool CameraWrapper::readProcessed(cv::Mat &src) {
 //	cerr << "Get-1" << endl;
-    if (CameraGetImageBuffer(h_camera, &frame_info, &pby_buffer, 1000) == CAMERA_STATUS_SUCCESS) {
+    if (CameraGetImageBuffer(h_camera, &frame_info, &pby_buffer, 500) == CAMERA_STATUS_SUCCESS) {
        CameraImageProcess(h_camera, pby_buffer, rgb_buffer,
                           &frame_info);  // this function is super slow, better not to use it.
        if (iplImage) {
@@ -192,9 +222,56 @@ bool CameraWrapper::readProcessed(cv::Mat &src) {
    }
 }

+bool CameraWrapper::readCallback(cv::Mat &src) {
+    timeval ts, te;
+    gettimeofday(&ts, NULL);
+    while(qtail==qhead){
+        gettimeofday(&te, NULL);
+        if(getTimeIntervalms(te, ts) > 500){
+            return false;
+        }
+    }
+
+    src = src_queue[qtail];
+//    cout << src.size << endl;
+    qtail = (qtail+1)%2;
+    return true;
+}
+
+bool CameraWrapper::readWithThread(cv::Mat &src){
+    if(readThread != nullptr){
+        readThread->join();
+        src = src_queue[qtail];
+        qtail = (qtail+1)%2;
+        delete readThread;
+        readThread = new std::thread([&](){
+            readProcessed(src_queue[qhead]);
+            qhead = (qhead+1)%2;
+        });
+    }else{
+        readThread = new std::thread([&](){
+            readProcessed(src_queue[qhead]);
+            qhead = (qhead+1)%2;
+        });
+        readThread->join();
+        src = src_queue[qtail];
+        qtail = (qtail+1)%2;
+        delete readThread;
+        readThread = new std::thread([&](){
+            readProcessed(src_queue[qhead]);
+            qhead = (qhead+1)%2;
+        });
+    }
+    return true;
+}
+
 CameraWrapper::~CameraWrapper() {
    CameraUnInit(h_camera);
    //注意，先反初始化后再free
    if (rgb_buffer != nullptr)
        free(rgb_buffer);
+    if(readThread != nullptr){
+        readThread->detach();
+        delete readThread;
+    }
 }
--- a/tools/TrainCNN/backward.py
+++ b/tools/TrainCNN/backward.py
@@ -1,6 +1,7 @@
 #!/usr/bin/python3
 print("Preparing...")
 import tensorflow as tf
+import os
 from tqdm import tqdm
 import generate
 import forward
@@ -35,28 +36,22 @@ def save_bias(fp, val):
        print(val[i], file=fp)


-def save_para(folder, paras):
-    with open(folder + "/conv1_w", "w") as fp:
-        save_kernal(fp, paras[0])
-    with open(folder + "/conv1_b", "w") as fp:
-        save_bias(fp, paras[1])
-    with open(folder + "/conv2_w", "w") as fp:
-        save_kernal(fp, paras[2])
-    with open(folder + "/conv2_b", "w") as fp:
-        save_bias(fp, paras[3])
-    with open(folder + "/fc1_w", "w") as fp:
-        save_weight_mat(fp, paras[4])
-    with open(folder + "/fc1_b", "w") as fp:
-        save_bias(fp, paras[5])
-    with open(folder + "/fc2_w", "w") as fp:
-        save_weight_mat(fp, paras[6])
-    with open(folder + "/fc2_b", "w") as fp:
-        save_bias(fp, paras[7])
+def save_para(folder, paras, names, info):
+    os.system("mkdir %s/%s" % (folder, info))
+    for para, name in zip(paras, names):
+        fp = open("%s/%s/%s" % (folder, info, name), "w")
+        if name[-1:] == "b":
+            save_bias(fp, para)
+        elif name[:2] == "fc":
+            save_weight_mat(fp, para)
+        elif name[:4] == "conv":
+            save_kernal(fp, para)
+        fp.close()


-STEPS = 60000
-BATCH = 50
-LEARNING_RATE_BASE  = 0.001
+STEPS = 50000
+BATCH = 30
+LEARNING_RATE_BASE  = 0.0005
 LEARNING_RATE_DECAY = 0.99
 MOVING_AVERAGE_DECAY = 0.99

@@ -65,7 +60,7 @@ def train(dataset, show_bar=False):
    x = tf.placeholder(tf.float32, [None, generate.SRC_ROWS, generate.SRC_COLS, generate.SRC_CHANNELS])
    y_= tf.placeholder(tf.float32, [None, forward.OUTPUT_NODES])
    keep_rate = tf.placeholder(tf.float32)
-    nodes, vars = forward.forward(x, 0.01)
+    nodes, vars, vars_name = forward.forward(x, 0.01)
    y = nodes[-1]

    ce  = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y, labels=tf.argmax(y_, 1))
@@ -101,12 +96,11 @@ def train(dataset, show_bar=False):

            _, loss_value, step = sess.run(
                [train_op, loss, global_step],
-                feed_dict={x: images_samples, y_: labels_samples, keep_rate:0.2}
+                feed_dict={x: images_samples, y_: labels_samples, keep_rate:0.4}
            )

-            if (i-1) % 100 == 0:
-                if (i-1) % 500 == 0:
-                    test_images, test_labels = dataset.sample_test_sets(5000)
+            if step % 500 == 0:
+                test_images, test_labels = dataset.sample_test_sets(6000)
                test_acc, output = sess.run([accuracy, y], feed_dict={x: test_images, y_: test_labels, keep_rate:1.0})
                output = np.argmax(output, axis=1)
                real = np.argmax(test_labels, axis=1)
@@ -115,7 +109,7 @@ def train(dataset, show_bar=False):
                    print("label: %d, precise: %f, recall: %f" %
                        (n, np.mean(real[output==n]==n), np.mean(output[real==n]==n)))

-                    train_images, train_labels = dataset.sample_train_sets(5000)
+                train_images, train_labels = dataset.sample_train_sets(6000)
                train_acc, output = sess.run([accuracy, y], feed_dict={x: train_images, y_: train_labels, keep_rate:1.0})
                output = np.argmax(output, axis=1)
                real = np.argmax(train_labels, axis=1)
@@ -124,12 +118,16 @@ def train(dataset, show_bar=False):
                    print("label: %d, precise: %f, recall: %f" %
                          (n, np.mean(real[output==n]==n), np.mean(output[real==n]==n)))
                print("\n")
-
-                bar.set_postfix({"loss": loss_value, "train_acc": train_acc, "test_acc": test_acc})
-
+                if train_acc >= 0.99 and test_acc >= 0.99:
                    vars_val = sess.run(vars)
-        save_para("/home/xinyang/Workspace/RM_auto-aim/tools/para", vars_val)
-        print("save done!")
+                    save_para(
+                        "model",
+                        vars_val,
+                        vars_name,
+                        "steps:%d-train_acc:%f-test_acc:%f" % (step, train_acc, test_acc)
+                    )
+                bar.set_postfix({"loss": loss_value, "train_acc": train_acc, "test_acc": test_acc})
+        # print("save done!")

        # pred = sess.run(y, feed_dict={x: test_images, keep_rate:1.0})

--- a/tools/TrainCNN/forward.py
+++ b/tools/TrainCNN/forward.py
@@ -29,16 +29,22 @@ def max_pool_2x2(x):
 CONV1_KERNAL_SIZE = 5

 # 第一层卷积输出通道数
-CONV1_OUTPUT_CHANNELS = 8
+CONV1_OUTPUT_CHANNELS = 4

 # 第二层卷积核大小
 CONV2_KERNAL_SIZE = 3

 # 第二层卷积输出通道数
-CONV2_OUTPUT_CHANNELS = 16
+CONV2_OUTPUT_CHANNELS = 6
+
+# 第三层卷积核大小
+CONV3_KERNAL_SIZE = 3
+
+# 第三层卷积输出通道数
+CONV3_OUTPUT_CHANNELS = 8

 # 第一层全连接宽度
-FC1_OUTPUT_NODES = 100
+FC1_OUTPUT_NODES = 50

 # 第二层全连接宽度（输出标签类型数）
 FC2_OUTPUT_NODES = 15
@@ -49,6 +55,7 @@ OUTPUT_NODES = FC2_OUTPUT_NODES

 def forward(x, regularizer=None, keep_rate=tf.constant(1.0)):
    vars = []
+    vars_name = []
    nodes = []

    conv1_w = get_weight(
@@ -57,7 +64,10 @@ def forward(x, regularizer=None, keep_rate=tf.constant(1.0)):
    conv1_b = get_bias([CONV1_OUTPUT_CHANNELS])
    conv1   = tf.nn.relu(tf.nn.bias_add(conv2d(x, conv1_w), conv1_b))
    pool1   = avg_pool_2x2(conv1)
+    print("conv1: ", conv1.shape)
+    print("pool1: ", pool1.shape)
    vars.extend([conv1_w, conv1_b])
+    vars_name.extend(["conv1_w", "conv1_b"])
    nodes.extend([conv1, pool1])

    conv2_w = get_weight(
@@ -66,27 +76,41 @@ def forward(x, regularizer=None, keep_rate=tf.constant(1.0)):
    conv2_b = get_bias([CONV2_OUTPUT_CHANNELS])
    conv2 = tf.nn.relu(tf.nn.bias_add(conv2d(pool1, conv2_w), conv2_b))
    pool2 = avg_pool_2x2(conv2)
+    print("conv2: ", conv2.shape)
+    print("pool2: ", pool2.shape)
    vars.extend([conv2_w, conv2_b])
+    vars_name.extend(["conv2_w", "conv2_b"])
    nodes.extend([conv2, pool2])

-    pool_shape = pool2.get_shape().as_list()
-    node = pool_shape[1] * pool_shape[2] * pool_shape[3]
-    reshaped = tf.reshape(pool2, [-1, node])
+    conv3_w = get_weight(
+        [CONV3_KERNAL_SIZE, CONV3_KERNAL_SIZE, CONV2_OUTPUT_CHANNELS, CONV3_OUTPUT_CHANNELS]
+    )
+    conv3_b = get_bias([CONV3_OUTPUT_CHANNELS])
+    conv3 = tf.nn.relu(tf.nn.bias_add(conv2d(pool2, conv3_w), conv3_b))
+    print("conv3: ", conv3.shape)
+    vars.extend([conv3_w, conv3_b])
+    vars_name.extend(["conv3_w", "conv3_b"])
+    nodes.extend([conv3])
+
+    conv_shape = conv3.get_shape().as_list()
+    node = conv_shape[1] * conv_shape[2] * conv_shape[3]
+    reshaped = tf.reshape(conv3, [-1, node])
    reshaped = tf.nn.dropout(reshaped, keep_rate)
+    print("reshaped: ", reshaped.shape)

    fc1_w = get_weight([node, FC1_OUTPUT_NODES], regularizer)
    fc1_b = get_bias([FC1_OUTPUT_NODES])
    fc1   = tf.nn.relu(tf.matmul(reshaped, fc1_w) + fc1_b)
-    fc1   = tf.nn.dropout(fc1, keep_rate)
    vars.extend([fc1_w, fc1_b])
+    vars_name.extend(["fc1_w", "fc1_b"])
    nodes.extend([fc1])

    fc2_w = get_weight([FC1_OUTPUT_NODES, FC2_OUTPUT_NODES], regularizer)
    fc2_b = get_bias([FC2_OUTPUT_NODES])
-#    fc2   = tf.nn.softmax(tf.matmul(fc1, fc2_w) + fc2_b)
    fc2 = tf.matmul(fc1, fc2_w) + fc2_b
    vars.extend([fc2_w, fc2_b])
+    vars_name.extend(["fc2_w", "fc2_b"])
    nodes.extend([fc2])

-    return nodes, vars
+    return nodes, vars, vars_name

--- a/tools/para/conv1_b
+++ b/tools/para/conv1_b
@@ -1,9 +1,5 @@
-8
-0.19897088
-2.2773967
-0.07212669
-0.22893764
-0.022769619
-0.9122422
-1.3221853
-0.21709026
+4
+0.30789205
+-0.043525748
+0.2050164
+0.74044687
--- a/tools/para/conv1_w
+++ b/tools/para/conv1_w
@@ -1,604 +1,304 @@
 3
-8
+4
 5
 5
-0.42449534
-0.14145629
-0.12018829
-0.4759054
-0.54840434
-0.28412497
-0.40357414
-0.6718681
-0.92518467
-0.68381476
-0.28161842
-0.5942867
-0.594364
-0.7217546
-0.77334726
-0.18300368
-0.3455078
-0.270947
-0.3452664
-0.3419373
-0.07475002
-0.23191033
-0.5056218
-0.5653023
-0.49549124
-0.356733
-0.70811594
-0.6640918
-0.5230208
-0.5567686
-0.34954152
-0.49958408
-0.31039435
-0.55455387
-0.3291702
-0.0030061502
-0.13863257
-0.40160847
-0.53651094
-0.24113898
-0.1733842
-0.2476929
-0.5121031
-0.47203135
-0.43864155
-0.058570273
-0.33658004
-0.6619966
-0.73419136
-0.7045188
-0.32072416
-0.330073
-0.55653447
-0.5957652
-0.50899994
-0.50545204
-0.7220086
-0.7438928
-0.7675144
-0.7756041
-0.7274903
-0.7188184
-0.88465214
-1.1385522
-0.87919647
-0.6987736
-0.9527178
-0.95726067
-0.88654906
-0.8920211
-0.63958967
-0.8072574
-0.8347145
-0.7488403
-0.6869278
-0.90990585
-0.6184357
-0.1404704
-0.20029725
-0.23383331
-0.5078964
-0.07829093
-0.41757903
-0.5164328
-0.5336988
-0.21864063
-0.07738885
-0.46755105
-0.6110745
-0.64779043
-0.17151009
-0.07808888
-0.014565518
-0.27556416
-0.21755305
-0.42132428
-0.4472937
-0.686964
-0.6123315
-0.36468038
-0.39503688
-0.7375632
-0.6933682
-0.59439814
-0.5431905
-0.4742321
-0.8823627
-0.77081895
-0.7913887
-0.5136633
-0.5808819
-0.80829126
-1.0471532
-0.6837362
-0.6456983
-0.6088025
-0.9599121
-0.79199773
-0.64847285
-0.7558641
-0.72989166
-0.8269554
-0.70485115
-0.773673
-0.5633863
-0.6651138
-0.6533556
-0.8055196
-0.70568013
-0.59960866
-0.38753736
-0.75367606
-0.6284249
-0.73716867
-0.4809728
-0.43523777
-0.5271899
-0.6199644
-0.6201511
-0.58342254
-0.41127473
-0.54275876
-0.53538835
-0.56108046
-0.48543084
-0.74164164
-0.64662766
-0.856863
-0.6305302
-0.5739134
-0.95437706
-1.2035962
-1.2448038
-1.1763076
-1.0001042
-0.78790265
-1.0511717
-1.0552691
-1.2410581
-0.8918281
-0.68448544
-0.8909238
-0.8172731
-0.8248441
-0.9374103
-0.545665
-0.65757316
-0.88965493
-0.78279305
-0.5069958
-0.4710934
-0.6540229
-0.7689269
-0.5900862
-0.6000688
-0.8985509
-0.5367204
-0.17939755
-0.1239284
-0.31574073
-0.29208726
-0.11087964
-0.13841438
-0.6425852
-0.6579574
-0.1677956
-0.5068273
-0.6169031
-0.46504426
-0.56336105
-0.007133658
-0.056750543
-0.083241865
-0.06789419
-0.14583983
-0.6158809
-0.87346137
-0.80676347
-0.70183337
-0.6015298
-1.5393821
-2.1624126
-2.5511134
-2.7846441
-2.368676
-2.4525952
-3.2989857
-4.029959
-3.827302
-3.2697773
-2.9329424
-3.9139073
-4.2881436
-4.3753977
-3.3441596
-2.6599689
-3.4512682
-3.6378286
-3.5350695
-2.7694979
-2.0715785
-2.4096649
-2.262437
-1.9068127
-1.4933991
-0.067793235
-0.4761605
-0.8775232
-0.79931116
-1.0089498
-0.3564602
-0.70624393
-1.1053402
-1.3865718
-1.4715241
-0.46418893
-1.059161
-1.4488351
-1.5534589
-1.6705322
-0.79103005
-1.1441072
-1.3800188
-1.5281351
-1.5740242
-0.9623409
-0.98793215
-1.1669405
-1.1152183
-1.2492605
-0.15674886
-0.038593877
-0.17274536
-0.09258732
-0.17494603
-0.105809644
-0.31649294
-0.343589
-0.53679705
-0.4734059
-0.3889172
-0.5807728
-0.67378217
-0.6827083
-0.7268035
-0.60358465
-0.7300566
-1.0874548
-1.0207677
-0.84341484
-0.60612875
-0.8210886
-1.0350081
-0.87764287
-1.0081718
-1.3859847
-2.0778801
-2.148741
-2.4518552
-1.9672226
-2.5095375
-3.0291762
-3.4696164
-3.2318525
-2.4367645
-2.8905299
-3.4577188
-3.8412285
-3.5718715
-2.7076635
-2.91653
-3.1068695
-3.2963896
-2.8063552
-1.9206675
-1.9905787
-2.0144224
-1.754742
-1.4237579
-0.83947563
-0.2254852
-0.03469266
-0.05631122
-0.04464834
-0.2178017
-0.13702668
-0.16909239
-0.26976383
-0.068580456
-0.12111518
-0.39565983
-0.57780766
-0.48672187
-0.3581179
-0.2925836
-0.5067063
-0.42480332
-0.46175724
-0.51814044
-0.094496034
-0.3766073
-0.46016395
-0.40080497
-0.3402491
-0.21284883
-0.13231997
-0.013396452
-0.083948776
-0.012691039
-0.11314143
-0.09580059
-0.2377315
-0.41866076
-0.15490921
-0.103149235
-0.18240465
-0.1797153
-0.4242559
-0.18161982
-0.2963119
-0.20458256
-0.20636415
-0.39781362
-0.23634791
-0.122939676
-0.24806741
-0.13085991
-0.16097488
-0.020796934
-0.17553945
-0.6122431
-0.5468711
-0.7881819
-0.70415086
-0.76530534
-0.23065606
-0.26207468
-0.22612737
-0.3477262
-0.33311552
-0.1145771
-0.064461164
-0.15550038
-0.17301485
-0.1755835
-0.46472925
-0.32503256
-0.30224133
-0.49536815
-0.5347998
-0.62774706
-0.52409613
-0.46520838
-0.47201794
-0.62468505
-1.2727304
-1.8184105
-2.2249503
-2.4140065
-1.7565204
-2.7623854
-3.196057
-3.6894598
-3.521049
-2.7936237
-3.4329123
-3.8600874
-4.2160697
-3.905414
-2.9514873
-3.2884743
-3.810133
-3.8212693
-3.4374819
-2.47835
-2.4106874
-2.5125678
-2.35327
-1.9101305
-1.2425265
-0.03470229
-0.7919308
-1.3452832
-1.2696229
-1.04937
-0.9344194
-1.8166728
-2.1765978
-2.4447913
-1.6805668
-1.2032535
-2.2395353
-2.6600883
-2.8381574
-1.9751589
-1.1395236
-1.8099883
-2.0537355
-2.181522
-1.5130651
-0.4666265
-0.76658946
-0.99030006
-0.7782916
-0.19490543
-0.86873436
-0.6494321
-0.3110133
-0.086757496
-0.061029676
-0.79375947
-0.47139648
-0.07530813
-0.17621484
-0.28143188
-0.54239106
-0.15787664
-0.18588926
-0.3305964
-0.52410334
-0.43658426
-0.16318789
-0.25725472
-0.3821112
-0.43296495
-0.34542954
-0.13750273
-0.1684193
-0.066175155
-0.109317034
-0.6713239
-0.704688
-0.41100273
-0.5969387
-0.5914036
-0.5237271
-0.36877418
-0.32102436
-0.2706671
-0.14425935
-0.24273401
-0.05300261
-0.0068080607
-0.030264616
-0.07868397
-0.11762654
-0.19707958
-0.21811731
-0.44130355
-0.21699134
-0.060512982
-0.19437213
-0.55949223
-0.32470936
-0.28635567
-1.1079576
-1.7189704
-2.413394
-2.4546094
-1.8035649
-2.1479077
-3.1008422
-3.601715
-3.365891
-2.4578235
-2.5486002
-3.5174582
-3.9265308
-3.515769
-2.5787847
-2.4205306
-3.1097507
-3.152673
-2.882059
-1.976877
-1.6673613
-1.9114503
-1.7297668
-1.4748011
-0.6597555
-0.42657572
-0.52120584
-0.68196577
-0.69425035
-0.9403177
-0.30822635
-0.46618125
-0.5267847
-0.493324
-0.65498495
-0.23576419
-0.16038808
-0.16629791
-0.3212767
-0.59757406
-0.004436309
-0.00716624
-0.11501519
-0.18837191
-0.39238173
-0.050632488
-0.14269532
-0.121495865
-0.26217306
-0.37235448
-0.6217619
-0.75724506
-0.81917787
-0.82933223
-0.8388242
-0.8166705
-0.95639604
-0.9749462
-1.0679555
-0.88015705
-0.9423948
-1.1189271
-0.9902488
-1.072058
-1.1399423
-0.96763176
-0.97359556
-1.1033723
-1.064764
-0.90642184
-0.70701605
-0.80866563
-0.71614605
-0.6529501
-0.8378095
-0.0864334
-0.05153916
-0.06473441
-0.12220235
-0.11894611
-0.2647296
-0.62840766
-0.621625
-0.4897769
-0.37764832
-0.6673036
-1.0537109
-1.1991285
-1.0963191
-0.96230215
-1.0695108
-1.3443365
-1.3382062
-1.6774666
-1.5912658
-1.3784939
-1.3979015
-1.6025708
-1.6357968
-1.8174925
-0.57204753
-1.1402864
-1.5789323
-1.6362445
-1.1193172
-1.8848814
-2.7911117
-3.185937
-3.0631964
-1.9419188
-2.6837914
-3.625602
-4.0801044
-3.6807868
-2.4854035
-2.679125
-3.3659801
-3.3164527
-3.082823
-1.911974
-1.7753979
-2.046517
-1.8074046
-1.2649333
-0.53168595
+-0.10729549
+0.09332296
+0.16133511
+-0.030649675
+0.06557881
+-0.053242918
+0.19213735
+0.15569621
+0.3419906
+0.1786888
+0.19507813
+0.24212381
+0.4231631
+0.08838954
+0.14048508
+0.15044038
+0.28268903
+0.13327996
+0.15814759
+-0.112561114
+0.056091413
+0.16789797
+-0.11966196
+0.038136892
+-0.14447927
+0.5679466
+0.48397887
+0.70058227
+0.5070652
+0.62130994
+0.5227163
+0.6784681
+0.5526853
+0.5210595
+0.396152
+0.31640542
+0.459686
+0.5702434
+0.5931808
+0.36256674
+0.3457865
+0.37771964
+0.4625998
+0.47415066
+0.3938988
+0.49735737
+0.26142278
+0.32298478
+0.3374469
+0.47159982
+-0.18274228
+-0.28124171
+-0.20628488
+-0.10543057
+-0.05156752
+-0.091269255
+-0.0054441793
+0.11289734
+-0.03441205
+-0.13315572
+0.20559654
+0.10798551
+-0.07475637
+-0.0047787298
+-0.22890174
+0.09388285
+0.059628867
+0.019401075
+0.05334341
+-0.1582362
+0.017811468
+-0.04889371
+-0.065829955
+-0.04788549
+-0.33305293
+-0.19116391
+-0.28003454
+-0.0820173
+-0.26577657
+-0.11826798
+-0.39622292
+-0.36510658
+-0.11750732
+-0.24734864
+-0.2803419
+-0.1519
+-0.23317525
+-0.007902931
+-0.16537969
+-0.19139121
+-0.023329364
+-0.30243063
+-0.24868572
+-0.029912082
+-0.23020941
+-0.31355116
+-0.11182503
+-0.0692445
+-0.35592338
+-0.09419684
+0.14571323
+0.5725029
+1.193062
+1.2903835
+0.9622877
+0.525414
+1.0394033
+1.4996175
+1.8389359
+1.3628953
+0.80069596
+1.4277213
+1.6703882
+1.7443689
+1.4166398
+0.6203115
+1.1946497
+1.7266799
+1.6697397
+1.3126082
+0.47843987
+0.89622074
+1.0277275
+1.0125589
+0.8202855
+0.5230944
+0.6464553
+0.5895753
+0.54708016
+0.26677945
+0.4320181
+0.59073323
+0.50776094
+0.50096244
+0.4625974
+0.49336708
+0.34545958
+0.56925744
+0.34867597
+0.16094568
+0.3861315
+0.5891766
+0.2174486
+0.20948303
+0.1000691
+0.21441981
+0.22367561
+0.19896027
+0.13936864
+0.03653052
+0.084119834
+0.3999692
+0.5730208
+0.6282694
+0.37436715
+0.46807006
+0.78820395
+0.96178275
+0.90761256
+0.87396187
+0.53964555
+0.9532658
+0.9813636
+1.0613359
+0.7933851
+0.5819338
+0.9753471
+1.2284292
+0.94796246
+0.82767236
+0.499911
+0.75207883
+0.68719465
+0.8145963
+0.6596423
+0.011669106
+0.18070999
+0.36343256
+0.3169534
+0.20783305
+0.03404153
+0.43459386
+0.36061177
+0.43950805
+0.38309076
+0.25803936
+0.4289157
+0.58764404
+0.58137804
+0.46296367
+0.1991764
+0.33899468
+0.64967453
+0.74104553
+0.57953227
+0.1818072
+0.32919815
+0.6142047
+0.47143123
+0.3723565
+-0.29113623
+0.08584255
+0.49546224
+0.80879575
+0.72197473
+-0.01818511
+0.4448289
+0.86975175
+1.0318508
+0.70635146
+0.2590701
+0.55703586
+1.2021183
+1.1355498
+0.929514
+-0.0021778075
+0.49526876
+0.8620847
+0.97750324
+0.80483395
+0.066819414
+0.0533782
+0.2378203
+0.48883578
+0.46825472
+-0.08690677
+-0.098846406
+-0.11066464
+-0.33117262
+-0.13379335
+-0.10680438
+-0.092536114
+-0.06396752
+-0.2411913
+-0.4162492
+-0.27065766
+-0.0127258655
+-0.22221713
+-0.10054428
+-0.25341535
+-0.36750633
+-0.0970442
+-0.28783646
+-0.3083135
+-0.37662375
+-0.24937427
+-0.36285818
+-0.32178807
+-0.41167092
+-0.64311564
+0.15358478
+0.45863125
+0.680853
+0.69516915
+0.73355865
+0.36617115
+0.8283799
+0.94936144
+0.96523696
+0.7651339
+0.4159632
+0.9592908
+1.038713
+1.1703031
+0.83268124
+0.62078
+0.8680032
+1.1751134
+1.0394844
+0.9290284
+0.527504
+0.75912774
+0.86471754
+0.89958185
+0.7007667
+0.11127112
+0.30424586
+0.6125697
+0.51516706
+0.4895497
+0.32818803
+0.44973958
+0.61502695
+0.57045287
+0.58641577
+0.28688067
+0.6754326
+0.85857826
+0.6959833
+0.67804474
+0.38325444
+0.5163973
+0.64538044
+0.79433715
+0.70066565
+0.38258222
+0.5113986
+0.61964786
+0.8566454
+0.7331395
--- a/tools/para/conv2_b
+++ b/tools/para/conv2_b
@@ -1,17 +1,7 @@
-16
-0.032270256
-2.3110154
-0.4078679
-0.30516425
-0.027841559
-0.80007833
-0.10999302
-0.83167756
-0.09058099
-0.67901427
-0.2736649
-0.08881351
-0.10975416
-0.25698876
-0.076515704
-0.017468728
+6
+-0.66018975
+-0.010440656
+1.4320896
+0.6933768
+-0.65163565
+0.5277429
--- a/tools/para/conv2_w
+++ b/tools/para/conv2_w
--- a/tools/para/fc1_b
+++ b/tools/para/fc1_b
@@ -1,101 +1,51 @@
-100
-0.39040318
-0.031096617
-0.06425226
-0.24911235
-0.002578787
-0.086705275
-0.0322658
-0.017816741
-0.11621032
-0.21196772
-0.004639828
-0.023076132
-0.50997764
-0.04299724
-0.01989839
-0.011238396
-0.003221448
-0.019384952
-0.0007764693
-0.015599826
-0.16373938
-0.0027049272
-0.18095633
-0.050923813
-0.12674743
-0.064153716
-0.028386148
-0.059802737
-0.036068685
-0.004065791
-0.03783843
-0.16458924
-0.0328307
-0.032716025
-0.020594684
-0.042352736
-0.084991984
-0.028080234
-0.001538593
-0.10711875
-0.024680987
-0.008004385
-0.5063542
-0.09158748
-0.08181085
-0.22574262
-0.075171836
-0.28233245
-0.024944687
-0.0029645876
-0.041441295
-0.08904015
-0.30993482
-0.06328518
-0.0075723003
-0.005151164
-0.0021952058
-0.013833341
-0.023337327
-0.01824665
-0.025177158
-0.067239
-0.02126352
-0.11769418
-0.64603645
-0.014887376
-0.14686602
-0.020528413
-0.018256638
-0.0017088759
-0.018110225
-0.003289471
-1.0441891
-0.30619404
-0.001282074
-0.09424017
-0.24455559
-0.026046017
-0.004658401
-0.022633847
-0.022873487
-0.4393057
-0.033948973
-0.042779494
-0.0059623853
-0.6859317
-0.19052452
-0.020080235
-0.010588832
-0.012147919
-0.002949453
-0.41500625
-0.16353038
-0.023607356
-0.38747007
-0.014272043
-0.0033837124
-0.1627222
-0.055108108
-0.74174875
+50
+-0.24530283
+-0.073897995
+-0.067673266
+-0.043941643
+0.339086
+-0.2782574
+-0.16835077
+-0.01852681
+0.3498275
+0.38015145
+-0.016264258
+-0.14133461
+0.24981335
+-0.025037237
+0.736614
+-0.1785015
+-0.040974453
+0.2144458
+0.23580065
+-0.115499295
+-0.09057627
+0.28634802
+0.58878446
+-0.12215623
+-0.10074399
+0.32165715
+0.39080995
+0.26881945
+-0.14676444
+0.50808805
+-0.053912967
+-0.111540824
+-0.10144225
+0.0035003617
+-0.010641405
+0.3912463
+0.06348127
+-0.11809033
+-0.024349451
+0.4927379
+0.23499253
+-0.042348947
+0.44671503
+0.23698664
+-0.18389338
+0.10498202
+-0.12768361
+-0.03513563
+0.14860639
+0.22698319
--- a/tools/para/fc1_w
+++ b/tools/para/fc1_w
--- a/tools/para/fc2_b
+++ b/tools/para/fc2_b
@@ -1,16 +1,16 @@
 15
-0.7158619
-0.074666105
-0.20262705
-0.32018387
-0.35891113
-0.021645868
-0.2373232
-0.17633525
-0.4311263
-0.2179705
-0.21293324
-0.20182535
-0.8007338
-0.02198011
-0.30222887
+0.45424584
+-0.028361967
+-0.06438486
+-0.15962061
+-0.20572336
+-0.111016475
+0.24467993
+-0.012314044
+-0.11952816
+-0.20754007
+-0.18928203
+-0.10316595
+-0.33859444
+0.15107821
+0.022365652
--- a/tools/para/fc2_w
+++ b/tools/para/fc2_w