Free-Shape Polygonal Object Localization

Abstract

Abstract. Polygonal objects are prevalent in man-made scenes. Early approaches to detecting them relied mainly on geometry while subsequent ones also incorporated appearance-based cues. It has recently been shown that this could be done fast by searching for cycles in graphs of line-fragments, provided that the cycle scoring function can be expressed as additive terms attached to individual fragments. In this paper, we propose an approach that eliminates this restriction. Given a weighted linefragment graph, we use its cyclomatic number to partition the graph into managebly-sized sub-graphs that preserve nodes and edges with a high weight and are most likely to contain object contours. Object contours are then detected as maximally scoring elementary circuits enumerated in each sub-graph. Our approach can be used with any cycle scoring function and multiple candidates that share line fragments can be found. This is unlike in other approaches that rely on a greedy approach to finding candidates. We demonstrate that our approach significantly outperforms the state-of-the-art for the detection of building rooftops in aerial images and polygonal object categories from ImageNet.

Fast Interactive Object Annotation With Curve-GCN

“…We analyze both automatic and interactive regimes, and compare to state-of-the-art baselines for both. For cross-domain experiments, we evaluate the generalization capability of our Cityscapes-trained model on the KITTI dataset [13] and four out-of-domain datasets, ADE20K [38], Aerial Rooftop [31], Cardiac MR [30], and ssTEM [14], following Polygon-RNN++ [2].…”

“…We follow [2] and use our Cityscapes-trained model and test it on KITTI [13] (in-domain driving dataset), ADE20k [38] (general scenes), Rooftop [31] (aerial imagery), and two medical datasets [30,16,14]. Quantitative Results.…”

“…We evaluate our Curve-GCN on the challenging Cityscapes dataset [10], where we outperform Polygon-RNN++ and PSP-Deeplab/DEXTR in both automatic and interactive settings. We also show that our model outperforms the baselines in cross-domain annotation, that is, a model trained on Cityscapes is used to annotate general scenes [38], aerial [31], and medical imagery [16,14]. Code is available: https://github.com/ fidler-lab/curve-gcn.…”

Efficient Interactive Annotation of Segmentation Datasets with Polygon-RNN++

“…Figure 20 shows a few qualitative examples of our interactive simulation on the Cityscapes dataset. The automatically predicted polygons are shown in the first column while the second column shows the result after a certain [10] and domain (general scenes [43], aerial [34], medical [16,33,11]). We emphasize that here we use our model trained on Cityscapes without any fine-tuning on these datasets.…”

Towards Large-Scale City Reconstruction from Satellites

“…Object polygonalization methods typically depart from the detection of line-segments which are then assembled into polygons. This second step can be done, for instance, by searching for cycles in a graph of line-segments [21], or by connecting line-segments with a gap filling strategy [22]. Grouping atomic regions [23] is also a possible approach, especially when the number of objects is high, and the input image is big.…”