arcface人脸特征提取不同人脸，计算的相似度都一样，是怎么回事呢？

[goldwater668]

余弦距离计算代码：
def cosine_distance(matrix1, matrix2):
matrix1_matrix2 = np.dot(matrix1, matrix2.transpose())
matrix1_norm = np.sqrt(np.multiply(matrix1, matrix1).sum(axis=1))
matrix1_norm = matrix1_norm[:, np.newaxis]
matrix2_norm = np.sqrt(np.multiply(matrix2, matrix2).sum(axis=1))
matrix2_norm = matrix2_norm[:, np.newaxis]
cosine_distance = np.divide(matrix1_matrix2, np.dot(matrix1_norm, matrix2_norm.transpose()))
return cosine_distance

是不是没有进行归一化呢，归一化有没有好的函数推荐呢？

[shouxieai]

a = np.random.rand(1, 8)
b = np.random.rand(1, 8)
cosine_distance = (a @ b.T) / (np.linalg.norm(a) * np.linalg.norm(b))

[goldwater668]

我按照你上述的公式：
train_feature=arcface.commit(train_crop,train_face.landmark).get()
test_feature=arcface.commit(test_crop,test_face.landmark).get()
dist = np.dot(train_feature,test_feature.transpose()) / (np.linalg.norm(train_feature) * np.linalg.norm(test_feature))
得出结果却是：
./face/recognize/2ys1.jpg
dist: [[1.0000001]]
dist: [[1.0000001]]
./face/recognize/2ys3.jpg
dist: [[1.0000001]]
dist: [[1.0000001]]
dist: [[1.0000001]]
dist: [[1.0000001]]
dist: [[1.0000001]]
dist: [[1.0000001]]
dist: [[1.0000002]]
dist: [[1.0000001]]
dist: [[1.0000001]]
./face/recognize/2ys5.jpg
dist: [[1.0000001]]
dist: [[1.0000001]]
dist: [[1.0000001]]
./face/recognize/2ys2.jpg
dist: [[1.0000001]]

[shouxieai]

提交了arcface案例，请参考：https://github.com/shouxieai/tensorRT_Pro/blob/main/python/test_arcface.py

[goldwater668]

@shouxieai AttributeError: 'trtpy.libtrtpyc.FaceBox' object has no attribute 'width'

[shouxieai]

重新编译后解决

[goldwater668]

@shouxieai arcface在进行特征提取的时候，为什么需要输入人脸关键点呢？retinaface中有一个函数crop_face_and_landmark裁剪的人脸很大，我想自己裁剪人脸，然后输入到arcface中进行特征提取，应该怎么做呢，之所以这样做，是因为发现提取有些同一个
人脸特征相似度相差很大