Add video test (#2)

* video test

* add minicaffe support

* Update README.md

cpp/CMakeLists.txt

@@ -7,3 +7,6 @@ include(mini-caffe/mini-caffe.cmake)
 
 add_executable(jfda main.cpp jfda.hpp jfda.cpp)
 target_link_libraries(jfda caffe ${OpenCV_LIBS})
+
+add_executable(video video.cpp jfda.hpp jfda.cpp)
+target_link_libraries(video caffe ${OpenCV_LIBS})

cpp/README.md

@@ -0,0 +1,4 @@
+CPP code for JFDA
+=================
+
+View video demo on YouTube [here](https://www.youtube.com/watch?v=Ee9UT9Zf98s).

cpp/convert.sh

@@ -0,0 +1,5 @@
+#!/usr/bin/env bash
+
+ffmpeg -i ./build/test.mp4 -f mp3 -vn ./build/test.mp3
+
+ffmpeg -i ./build/test.mp3 -i ./build/result.avi -f mp4 -y ./build/result.mp4

cpp/video.cpp

@@ -0,0 +1,53 @@
+#include <iostream>
+#include "./jfda.hpp"
+#include "caffe/caffe.hpp"
+#include <opencv2/core/core.hpp>
+#include <opencv2/highgui/highgui.hpp>
+
+using namespace cv;
+using namespace std;
+using namespace jfda;
+
+int main(int argc, char* argv[]) {
+  VideoCapture cap("./test.mp4");
+  if (!cap.isOpened()) {
+    cout << "Can\'t open test.mp4" << endl;
+    return 0;
+  }
+  double rate = cap.get(CV_CAP_PROP_FPS);
+  double width = cap.get(CV_CAP_PROP_FRAME_WIDTH);
+  double height = cap.get(CV_CAP_PROP_FRAME_HEIGHT);
+  int fourcc = CV_FOURCC('X', 'V', 'I', 'D');
+  VideoWriter writer("./result.avi", fourcc, rate, Size(width, height), true);
+  string proto_dir = "../../proto/";
+  string model_dir = "../../model/";
+  JfdaDetector detector(proto_dir + "p.prototxt", model_dir + "p.caffemodel",
+                        proto_dir + "r.prototxt", model_dir + "r.caffemodel",
+                        proto_dir + "o.prototxt", model_dir + "o.caffemodel",
+                        proto_dir + "l.prototxt", model_dir + "l.caffemodel", 0);
+  detector.SetMinSize(24);
+  Mat frame;
+  caffe::Profiler* profiler = caffe::Profiler::Get();
+  while (cap.read(frame)) {
+    int64_t t1 = profiler->Now();
+    vector<FaceInfo> faces = detector.Detect(frame);
+    int64_t t2 = profiler->Now();
+    double fps = 1000000 / (t2 - t1);
+    for (int i = 0; i < faces.size(); i++) {
+      FaceInfo& face = faces[i];
+      cv::rectangle(frame, face.bbox, Scalar(0, 0, 255), 2);
+      for (int j = 0; j < 5; j++) {
+        cv::circle(frame, face.landmark5[j], 2, Scalar(0, 255, 0), -1);
+      }
+    }
+    char buff[30];
+    sprintf(buff, "%.2f FPS", fps);
+    cv::putText(frame, buff, Point2f(0, 10), FONT_HERSHEY_PLAIN, 1, Scalar(0, 255, 0));
+    writer.write(frame);
+    cv::imshow("frame", frame);
+    if (waitKey(1) >= 0) {
+      break;
+    }
+  }
+  return 0;
+}

demo.py

@@ -1,4 +1,5 @@
 #!/usr/bin/env python2.7
+# pylint: disable=bad-indentation, no-member, invalid-name, line-too-long
 
 import os
 import argparse
@@ -12,7 +13,11 @@ def main(args):
          'proto/r.prototxt', 'model/r.caffemodel',
          'proto/o.prototxt', 'model/o.caffemodel',
          'proto/l.prototxt', 'model/l.caffemodel',]
-  detector = JfdaDetector(net)
+  if args.minicaffe:
+    from jfda.mdetector import MiniCaffeDetector
+    detector = MiniCaffeDetector(net)
+  else:
+    detector = JfdaDetector(net)
   if args.pnet_single:
     detector.set_pnet_single_forward(True)
   param = {
@@ -63,6 +68,7 @@ def gen(img, bboxes, outname):
   parser = argparse.ArgumentParser()
   parser.add_argument('--gpu', type=int, default=-1, help='gpu id to use, -1 for cpu')
   parser.add_argument('--pnet-single', action='store_true', help='pnet use single forward')
+  parser.add_argument('--minicaffe', action='store_true', help='use minicaffe')
   args = parser.parse_args()
 
   if args.gpu >= 0:

fddb.py

@@ -1,4 +1,5 @@
 #!/usr/bin/env python2.7
+# pylint: disable=bad-indentation, no-member, invalid-name, line-too-long
 
 import argparse
 import cv2

jfda/detector.py

@@ -95,6 +95,9 @@ def detect(self, img, ths, min_size, factor, debug=False):
       data[i] = cv2.resize(face, (24, 24)).transpose((2, 0, 1))
     data = (data - 128) / 128
     prob, bbox_pred, landmark_pred = self._forward(self.rnet, data, ['prob', 'bbox_pred', 'landmark_pred'])
+    prob = prob.reshape(n, 2)
+    bbox_pred = bbox_pred.reshape(n, 4)
+    landmark_pred = landmark_pred.reshape(n, 10)
     keep = prob[:, 1] > ths[1]
     bboxes = bboxes[keep]
     bboxes[:, 4] = prob[keep, 1]
@@ -122,6 +125,9 @@ def detect(self, img, ths, min_size, factor, debug=False):
       data[i] = cv2.resize(face, (48, 48)).transpose((2, 0, 1))
     data = (data - 128) / 128
     prob, bbox_pred, landmark_pred = self._forward(self.onet, data, ['prob', 'bbox_pred', 'landmark_pred'])
+    prob = prob.reshape(n, 2)
+    bbox_pred = bbox_pred.reshape(n, 4)
+    landmark_pred = landmark_pred.reshape(n, 10)
     keep = prob[:, 1] > ths[2]
     bboxes = bboxes[keep]
     bboxes[:, 4] = prob[keep, 1]
@@ -158,6 +164,7 @@ def detect(self, img, ths, min_size, factor, debug=False):
         data[i, (3*j):(3*j+3)] = patch
     data = (data - 128) / 128
     offset = self._forward(self.lnet, data, ['landmark_offset'])[0]
+    offset = offset.reshape(n, 10)
     offset *= l.reshape((-1, 1))
     bboxes[:, 9:] += offset
     timer.toc()

jfda/mdetector.py

@@ -0,0 +1,19 @@
+# pylint: disable=bad-indentation, no-member, invalid-name, line-too-long
+from .detector import JfdaDetector
+import minicaffe as caffe
+
+
+class MiniCaffeDetector(JfdaDetector):
+
+    def __init__(self, nets):
+        assert len(nets) in [2, 4, 6, 8], 'wrong number of nets'
+        self.pnet, self.rnet, self.onet, self.lnet = None, None, None, None
+        if len(nets) >= 2:
+            self.pnet = caffe.Net(nets[0], nets[1])
+        if len(nets) >= 4:
+            self.rnet = caffe.Net(nets[2], nets[3])
+        if len(nets) >= 6:
+            self.onet = caffe.Net(nets[4], nets[5])
+        if len(nets) >= 8:
+            self.lnet = caffe.Net(nets[6], nets[7])
+        self.pnet_single_forward = False