SOLD2#
Line detection Line matching MIT
SOLD² detects line segments and describes them in one network, so lines can be matched across views the way
keypoints are. SOLD2 returns the segments of each image together with a dense descriptor
map, and its match method pairs the segments of two images. SOLD2_detector is the
detector alone.
Run it#
import torch
from kornia.color import rgb_to_grayscale
from kornia.feature import SOLD2
from kornia.io import load_image
img1 = load_image("church_1.png")[None] # (1, 3, H, W) float in [0, 1]
img2 = load_image("church_2.png")[None]
gray = rgb_to_grayscale(torch.cat([img1, img2])) # (2, 1, H, W)
sold2 = SOLD2(pretrained=True).eval()
with torch.no_grad():
out = sold2(gray)
lines1, lines2 = out["line_segments"] # list of (N, 2, 2) endpoints in (y, x) pixels, one per image
desc = out["dense_desc"] # (2, 128, H/4, W/4)
matches = sold2.match(lines1, lines2, desc[0:1], desc[1:2]) # (N1,) index into lines2, -1 = no match
matched1, matched2 = lines1[matches != -1], lines2[matches[matches != -1]]
Segments detected in each image (left, centre) and the matched pairs, one colour per pair (right).#
The input is a grayscale batch; out also carries the raw junction_heatmap and line_heatmap maps. The
detection and matching thresholds are set through the config argument of SOLD2.
Paper#
SOLD²: Self-supervised Occlusion-aware Line Description and Detection
Abstract: Compared to feature point detection and description, detecting and matching line segments offer additional challenges. Yet, line features represent a promising complement to points for multi-view tasks. Lines are indeed well-defined by the image gradient, frequently appear even in poorly textured areas and offer robust structural cues. We thus hereby introduce the first joint detection and description of line segments in a single deep network. Thanks to a self-supervised training, our method does not require any annotated line labels and can therefore generalize to any dataset. Our detector offers repeatable and accurate localization of line segments in images, departing from the wireframe parsing approach. Leveraging the recent progresses in descriptor learning, our proposed line descriptor is highly discriminative, while remaining robust to viewpoint changes and occlusions. We evaluate our approach against previous line detection and description methods on several multi-view datasets created with homographic warps as well as real-world viewpoint changes. Our full pipeline yields higher repeatability, localization accuracy and matching metrics, and thus represents a first step to bridge the gap with learned feature points methods.
Tasks: Line detection, Line description, Line matching
Datasets: Wireframe, YorkUrban, ETH3D
Conference: CVPR 2021
Licence: MIT
