处理本地视频和传播的问题 #71
Description
在数据生成过程中，对于陌生视频，总是会出现两个问题，第一个问题是./process_videos.sh video_list.txt产生的24个文件夹的伪标签部分有时候为空的问题，是因为某些匹配器效果不好吗？但是propagate 过程中又expected 24个 第二个问题是尽管生成了24个文件夹的伪标签，但是在propagate 过程中出现没有传递下去的情况（也就是这里的continue1,continue2,continue3出现全为False的情况），我想问的是针对这种情况，能不能只采用部分匹配器的结果进行传播呢，您在处理在线视频的时候遇到这种情况如何处理呢？按理说很多在线视频都存在匹配器效果不好的情况，本身匹配到的数量就少，传播下去的更少，如何保证传播过程中至少存在匹配的情况呢 continue1 = np.array([x in ids1[:, 0] for x in (ids1[:, 0] + self.skips[-1] * 1)])
print("continue1 shape = ",continue1.shape)
ids2 = reduce(intersected, idxs[self.skips[-2]])
print("ids2 shape = ",ids2.shape)
continue2 = np.array([x in ids2[:, 0] for x in ids1[:, 0]])
print("continue2 shape = {} | later shape = {}".format(continue2.shape,np.array([x in ids2[:, 0] for x in (ids1[:, 0] + self.skips[-2] * 1)]).shape))
continue2 = continue2 & np.array([x in ids2[:, 0] for x in (ids1[:, 0] + self.skips[-2] * 1)])
ids3 = reduce(intersected, idxs[self.skips[-3]])
continue3 = np.array([x in ids3[:, 0] for x in ids1[:, 0]])
continue3 = continue3 & np.array([x in ids3[:, 0] for x in (ids1[:, 0] + self.skips[-3] * 1)])
continue3 = continue3 & np.array([x in ids3[:, 0] for x in (ids1[:, 0] + self.skips[-3] * 2)])
continue3 = continue3 & np.array([x in ids3[:, 0] for x in (ids1[:, 0] + self.skips[-3] * 3)])
continues = continue1 & continue2 & continue3