Dense extreme inception network for edge detection

Edge detection is the basis of many computer vision applications. State art predominantly relies on deep learning with two decisive factors: dataset content and network architecture. Most publicly available datasets are not curated for edge tasks. Here, we address this limitation. First, argue that edges, contours boundaries, despite their overlaps, three distinct visual features requiring separate benchmark datasets. To end, present a new edges. Second, propose novel architecture, termed Dense Extreme Inception Network Detection (DexiNed), can be trained from scratch without any pre-trained weights. DexiNed outperforms other algorithms in presented dataset. It also generalizes well to fine-tuning. The higher quality perceptually evident thanks sharper finer edges it outputs.