File size: 2,473 Bytes
0a82b18
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
import torch
from pathlib import Path
import gdown
from copy import deepcopy
import torchvision.transforms as tfm


from matching import WEIGHTS_DIR, THIRD_PARTY_DIR, BaseMatcher
from matching.utils import to_numpy, resize_to_divisible, add_to_path

add_to_path(THIRD_PARTY_DIR.joinpath("EfficientLoFTR"), insert=0)

from src.loftr import LoFTR, full_default_cfg, opt_default_cfg, reparameter


class EfficientLoFTRMatcher(BaseMatcher):
    weights_src = "https://drive.google.com/file/d/1jFy2JbMKlIp82541TakhQPaoyB5qDeic/view"
    model_path = WEIGHTS_DIR.joinpath("eloftr_outdoor.ckpt")
    divisible_size = 32

    def __init__(self, device="cpu", cfg="full", **kwargs):
        super().__init__(device, **kwargs)

        self.precision = kwargs.get("precision", self.get_precision())

        self.download_weights()

        self.matcher = LoFTR(config=deepcopy(full_default_cfg if cfg == "full" else opt_default_cfg))

        self.matcher.load_state_dict(torch.load(self.model_path, map_location=torch.device("cpu"))["state_dict"])
        self.matcher = reparameter(self.matcher).to(self.device).eval()

    def get_precision(self):
        return "fp16"

    def download_weights(self):
        if not Path(EfficientLoFTRMatcher.model_path).is_file():
            print("Downloading eLoFTR outdoor... (takes a while)")
            gdown.download(
                EfficientLoFTRMatcher.weights_src,
                output=str(EfficientLoFTRMatcher.model_path),
                fuzzy=True,
            )

    def preprocess(self, img):
        _, h, w = img.shape
        orig_shape = h, w
        img = resize_to_divisible(img, self.divisible_size)
        return tfm.Grayscale()(img).unsqueeze(0), orig_shape

    def _forward(self, img0, img1):
        img0, img0_orig_shape = self.preprocess(img0)
        img1, img1_orig_shape = self.preprocess(img1)

        batch = {"image0": img0, "image1": img1}
        if self.precision == "mp" and self.device == "cuda":
            with torch.autocast(enabled=True, device_type="cuda"):
                self.matcher(batch)
        else:
            self.matcher(batch)

        mkpts0 = to_numpy(batch["mkpts0_f"])
        mkpts1 = to_numpy(batch["mkpts1_f"])

        H0, W0, H1, W1 = *img0.shape[-2:], *img1.shape[-2:]
        mkpts0 = self.rescale_coords(mkpts0, *img0_orig_shape, H0, W0)
        mkpts1 = self.rescale_coords(mkpts1, *img1_orig_shape, H1, W1)

        return mkpts0, mkpts1, None, None, None, None