diff --git a/tests/openface_api_tests.py b/tests/openface_api_tests.py index 75bfee3..419eafa 100644 --- a/tests/openface_api_tests.py +++ b/tests/openface_api_tests.py @@ -37,43 +37,14 @@ lfwSubset = os.path.join(openfaceDir, 'data', 'lfw-subset') dlibFacePredictor = os.path.join(dlibModelDir, "shape_predictor_68_face_landmarks.dat") -nn4_v1_model = os.path.join(openfaceModelDir, 'nn4.v1.t7') -nn4_v2_model = os.path.join(openfaceModelDir, 'nn4.v2.t7') +model = os.path.join(openfaceModelDir, 'nn4.small2.v1.t7') imgDim = 96 align = openface.AlignDlib(dlibFacePredictor) -nn4_v1 = openface.TorchNeuralNet(nn4_v1_model, imgDim=imgDim) -nn4_v2 = openface.TorchNeuralNet(nn4_v2_model, imgDim=imgDim) +net = openface.TorchNeuralNet(model, imgDim=imgDim) -def test_v1_pipeline(): - imgPath = os.path.join(exampleImages, 'lennon-1.jpg') - bgrImg = cv2.imread(imgPath) - if bgrImg is None: - raise Exception("Unable to load image: {}".format(imgPath)) - rgbImg = cv2.cvtColor(bgrImg, cv2.COLOR_BGR2RGB) - # assert np.isclose(norm(rgbImg), 11.1355) - - bb = align.getLargestFaceBoundingBox(rgbImg) - assert bb.left() == 341 - assert bb.right() == 1006 - assert bb.top() == 193 - assert bb.bottom() == 859 - - # Should be INNER_EYES_AND_BOTTOM_LIP by default. - alignedFace = align.align(imgDim, rgbImg, bb) - # assert np.isclose(norm(alignedFace), 8.30662) - - alignedFace_alt = align.align(imgDim, rgbImg, bb, - landmarkIndices=openface.AlignDlib.INNER_EYES_AND_BOTTOM_LIP) - assert np.isclose(norm(alignedFace), norm(alignedFace_alt)) - - rep = nn4_v1.forward(alignedFace) - cosDist = scipy.spatial.distance.cosine(rep, np.ones(128)) - assert np.isclose(cosDist, 1.01339430746) - - -def test_v2_pipeline(): +def test_pipeline(): imgPath = os.path.join(exampleImages, 'lennon-1.jpg') bgrImg = cv2.imread(imgPath) if bgrImg is None: @@ -91,6 +62,7 @@ def test_v2_pipeline(): landmarkIndices=openface.AlignDlib.OUTER_EYES_AND_NOSE) # assert np.isclose(norm(alignedFace), 7.61577) - rep = nn4_v2.forward(alignedFace) + rep = net.forward(alignedFace) cosDist = scipy.spatial.distance.cosine(rep, np.ones(128)) - assert np.isclose(cosDist, 0.981229293936) + print(cosDist) + assert np.isclose(cosDist, 0.938840385931)