lfw roc: Make sure every representation exists.

evaluation/lfw-roc.py

@@ -38,10 +38,12 @@ def main():
     args = parser.parse_args()
 
     print("Loading embeddings.")
-    paths = pd.read_csv("{}/labels.csv".format(args.workDir)).as_matrix()[:,1]
+    fname = "{}/labels.csv".format(args.workDir)
+    paths = pd.read_csv(fname, header=None).as_matrix()[:,1]
     paths = map(os.path.basename, paths) # Get the filename.
     paths = map(lambda path: os.path.splitext(path)[0], paths) # Remove the extension.
-    rawEmbeddings = pd.read_csv("{}/reps.csv".format(args.workDir)).as_matrix()
+    fname = "{}/reps.csv".format(args.workDir)
+    rawEmbeddings = pd.read_csv(fname, header=None).as_matrix()
     embeddings = dict(zip(*[paths, rawEmbeddings]))
 
     pairs = loadPairs()
@@ -89,19 +91,20 @@ def analyze_accuracy(workDir, pairs, embeddings):
                 else:
                     raise Exception(
                         "Unexpected pair length: {}".format(len(pair)))
+
                 # print(name1,name2)
-                if name1 not in embeddings or name2 not in embeddings:
-                    # Representation of one or both people is not available
-                    # since sometimes the face cannot be aligned.
-                    # Guess they are the same and note the error.
-                    num_errors += 1
-                    predict_same = True
-                else:
-                    vec1 = embeddings[name1]
-                    vec2 = embeddings[name2]
-                    diff = vec1-vec2
-                    dist = np.dot(diff.T, diff)
-                    predict_same = dist < threshold
+                if name1 not in embeddings:
+                    print('Error: Representation for {} not found.'.format(name1))
+                    sys.exit(-1)
+                if name2 not in embeddings:
+                    print('Error: Representation for {} not found.'.format(name2))
+                    sys.exit(-1)
+
+                vec1 = embeddings[name1]
+                vec2 = embeddings[name2]
+                diff = vec1-vec2
+                dist = np.dot(diff.T, diff)
+                predict_same = dist < threshold
 
                 if predict_same and actual_same: tp += 1
                 elif predict_same and not actual_same: fp += 1
@@ -134,7 +137,7 @@ def plot_accuracy(workDir):
     openbrData['Y'] = 1-openbrData['Y']
     openbrData.plot(x='X', y='Y', legend=True, ax=ax)
 
-    ax.legend(['FaceNet', 'OpenBR v1.0.0'], loc='lower right')
+    ax.legend(['FaceNet nn4.v1', 'OpenBR v1.0.0'], loc='lower right')
 
     plt.plot([0,1], color='k', linestyle='dashed')