results

Author: Shashank Tyagi

main_prediction.py

@@ -1,6 +1,6 @@
 import tensorflow as tf
 import os
-from model import *
+from model_prediction import *
 
 
 

model_prediction.py

@@ -206,7 +206,7 @@ def network(self,inputs,reuse=False):
 			conv_all = slim.conv2d(concat_feat, 192, [1,1], 1, padding= 'VALID', scope='conv_all')
 
 			shape = int(np.prod(conv_all.get_shape()[1:]))
-			fc_full = slim.fully_connected(tf.reshape(conv_all, [-1, shape]), 3072, scope='fc_full')
+			fc_full = slim.fully_connected(tf.reshape(tf.transpose(conv_all, [0,3,1,2]), [-1, shape]), 3072, scope='fc_full')
 
 			fc_detection = slim.fully_connected(fc_full, 512, scope='fc_detection1')
 			fc_landmarks = slim.fully_connected(fc_full, 512, scope='fc_landmarks1')
@@ -239,7 +239,6 @@ def predict(self):
 				print count
 				batch_imgs, batch_labels, batch_landmarks, batch_visibility, batch_pose, batch_gender = self.sess.run([self.images,self.labels,self.land, self.vis, self.po, self.gen])
 				batch_imgs = (batch_imgs - 127.5) / 128.0
-				# input_feed={self.X: batch_imgs, self.detection: batch_labels, self.landmarks: batch_landmarks, self.visibility: batch_visibility, self.pose: batch_pose, self.gender: np.squeeze(batch_gender)}
 
 				net_preds = self.sess.run(self.net_output, feed_dict={self.X: batch_imgs})
 				result.append(np.concatenate(net_preds, axis=1))

results_analysis.py

@@ -0,0 +1,42 @@
+import numpy as np
+np.set_printoptions(linewidth=200)
+from sklearn.metrics import roc_curve, precision_recall_curve
+import matplotlib.pyplot as plt
+
+
+def softmax(x):
+    return np.exp(x) / np.sum(np.exp(x), axis=1, keepdims=True)
+
+
+pred = np.load('test_results.npy')
+truth = np.load('truth.npy')
+
+print pred.shape, truth.shape
+
+
+prob = softmax(pred[:,:2])
+# print prob.sum(axis = 1)
+print pred
+
+
+print 'Detection accuracy: ', np.sum(np.argmax(prob, axis = 1) == truth[:,0].astype(np.bool))/float(prob.shape[0])
+
+print 'Gender accuracy: ', np.sum(np.argmax(softmax(pred[:,68:70]), axis = 1) == truth[:,-1].astype(np.bool))/float(prob.shape[0])
+
+# detection
+fpr, tpr, thresholds = roc_curve(truth[:,0], prob[:,1])
+precision, recall, th = precision_recall_curve(truth[:,0], prob[:,1])
+
+# plt.plot(recall, precision)
+# plt.xlabel('Recall')
+# plt.ylabel('Precision')
+# # plt.savefig('detection_pr_re.eps', format='eps', dpi=1000)
+# plt.show()
+
+
+# plt.plot(fpr, tpr)
+# plt.xlabel('False positive rate')
+# plt.ylabel('True positive rate')
+# # plt.savefig('detection_roc.eps', format='eps', dpi=1000)
+# plt.show()
+