fx = np.arange(-np.pi, np.pi, 3 * np.pi / 180)
features = np.zeros((pts_num, len(fx)))
for i in range(k):
sigma = 2.5 * np.exp(-rho_feature[:, i] * 100) + .04
tmp = np.exp(-np.subtract.outer(theta_feature[:, i], fx) ** 2 / 2 / sigma[:, np.newaxis] ** 2)
tmp = tmp * (vol_feature[:, i] * rho_feature[:, i] ** 2 / sigma)[:, np.newaxis]
features += tmp
features = features / np.sqrt(np.sum(features ** 2, axis=1)).reshape((pts_num, 1))
请问一下在哪里可以找到feature descriptor的解释？看上去与正态分布的函数很相近但是看不太明白。谢谢。