基于深度学习Superpoint 的Python图像全景拼接

参考
https://github.com/kushalvyas/Python-Multiple-Image-Stitching
https://github.com/MagicLeapResearch/SuperPointPretrainedNetwork
https://github.com/syinari0123/SuperPoint-VO
用superpoint方法代替surf提取图像特征，进行Python版本的图像拼接。
注意，Python版本图像拼接效果并不好，本博客只是学习记录。
改动后的matchers.py如下：

import cv2
import numpy as np 
from sp_extractor import SuperPointFrontend
class matchers:
	def __init__(self):
		self.surf = cv2.xfeatures2d.SURF_create()
                self.detector = SuperPointFrontend(weights_path="superpoint_v1.pth",
                                           nms_dist=4,
                                           conf_thresh=0.015,
                                           nn_thresh=0.7,
                                           cuda=True)
		FLANN_INDEX_KDTREE = 0
		index_params = dict(algorithm=0, trees=5)
		search_params = dict(checks=50)
		self.flann = cv2.FlannBasedMatcher(index_params, search_params)

	def match(self, i1, i2, direction=None):
		imageSet1 = self.getSURFFeatures(i1)
		imageSet2 = self.getSURFFeatures(i2)
		print "Direction : ", direction
		
                matches = self.flann.knnMatch(
			np.asarray(imageSet2['des'],np.float32),
			np.asarray(imageSet1['des'],np.float32),
			k=2
			)
		good = []
		for i , (m, n) in enumerate(matches):
			if m.distance < 0.7*n.distance:
				good.append((m.trainIdx, m.queryIdx))

		if len(good) > 4:
			pointsCurrent = imageSet2['kp']
			pointsPrevious = imageSet1['kp']

			matchedPointsCurrent = np.float32(
				[pointsCurrent[i] for (__, i) in good]
			)
			matchedPointsPrev = np.float32(
				[pointsPrevious[i] for (i, __) in good]
				)

			H, s = cv2.findHomography(matchedPointsCurrent, matchedPointsPrev, cv2.RANSAC, 4)
			return H
		return None

	def getSURFFeatures(self, im):
		gray = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
                pts, desc, heatmap = self.detector.run(gray)
		#kp, des = self.surf.detectAndCompute(gray, None)
		#superpoint的pts和desc格式为3*n,改为n*2。
                pts=np.delete(pts,2,axis=0)
                desc=np.delete(desc,2,axis=0)
                pts=np.transpose(pts)
                desc=np.transpose(desc)
                pts=pts.tolist()#矩阵转换为list,
                desc=desc.tolist()
                
                return {'kp':pts, 'des':desc}
                ```