The retrieval speed

[henry32144]

Hi, Thanks for your awesome implementation. This repository really helps me a lot

I am trying to make an image retrieval system with 53 images in my database for testing. I found that in the PQ search step, the system runs fast. However, the system getting slower in the geometric verification step (about 5 secs for one search). Currently, I apply a limit to the get_ransac_result function as the snippet below. And I am wondering if there are other ways to speed up the image retrieval system without losing too much precision?

def get_ransac_result(query_img2imgFreq, top_freq=10):
    query_inlier_rank = {}

    # MODIFY: pick only top similar images to get faster speed
    top_freq = 10

    # explore each image's frequency-based ranked image indices
    for query_i in tqdm(range(len(query_img2imgFreq))):

        # MODIFY: apply the limit here
        ranked_list = query_img2imgFreq[query_i]['index'][:top_freq]
        ...

Thank you very much.