opensfm/bow.py

import os.path

import cv2
import numpy as np
from opensfm import context


class BagOfWords:
    def __init__(self, words, frequencies):
        self.words = words
        self.frequencies = frequencies
        self.weights = np.log(frequencies.sum() / frequencies)
        FLANN_INDEX_KDTREE = 1
        flann_params = {"algorithm": FLANN_INDEX_KDTREE, "trees": 8, "checks": 300}
        self.index = context.flann_Index(words, flann_params)

    def map_to_words(self, descriptors, k, matcher_type="FLANN"):
        if matcher_type == "FLANN":
            params = {"checks": 200}
            idx, dist = self.index.knnSearch(descriptors, k, params=params)
        else:
            matcher = cv2.DescriptorMatcher_create(matcher_type)
            matches = matcher.knnMatch(descriptors, self.words, k=k)
            idx = [[int(n.trainIdx) for n in m] for m in matches]
            idx = np.array(idx).astype(np.int32)
        return idx

    def histogram(self, words):
        h = np.bincount(words, minlength=len(self.words)) * self.weights
        return h / h.sum()

    def bow_distance(self, w1, w2, h1=None, h2=None):
        if h1 is None:
            h1 = self.histogram(w1)
        if h2 is None:
            h2 = self.histogram(w2)
        return np.fabs(h1 - h2).sum()


def load_bow_words_and_frequencies(config):
    if config["bow_file"] == "bow_hahog_root_uchar_10000.npz":
        assert config["feature_type"] == "HAHOG"
        assert config["feature_root"]
        assert config["hahog_normalize_to_uchar"]

    bow_file = os.path.join(context.BOW_PATH, config["bow_file"])
    bow = np.load(bow_file)
    return bow["words"], bow["frequencies"]


def load_vlad_words_and_frequencies(config):
    if config["vlad_file"] == "bow_hahog_root_uchar_64.npz":
        assert config["feature_type"] == "HAHOG"
        assert config["feature_root"]
        assert config["hahog_normalize_to_uchar"]

    vlad_file = os.path.join(context.BOW_PATH, config["vlad_file"])
    vlad = np.load(vlad_file)
    return vlad["words"], vlad["frequencies"]


def load_bows(config) -> BagOfWords:
    words, frequencies = load_bow_words_and_frequencies(config)
    return BagOfWords(words, frequencies)