search

cvg / DeepLSD

Implementation of the paper "DeepLSD: Line Segment Detection and Refinement with Deep Image Gradients"
MIT License
486 stars 62 forks source link

How to optimise DeepLSD with ortho-image ? #20

Closed aquexexl closed 1 year ago

aquexexl commented 1 year ago

Hi guys! Thank you for your work ! I work with image from scan 3D like cut or front of a building 18045-fac-principale_inf (6) Is it possible to work and optimize this type of image with no perspective or angle ? Do I have to change some parameters ? Thank you !

rpautrat commented 1 year ago

Hi, at test time you only need a single image as input. So it should work fine with the default parameters I think, if your images are similar to the ones you show here.

aquexexl commented 1 year ago

Okay thank you for your answer. Do I have to change some parameter for the training, I have created a new wireframe dataset with my own images. I had to cut my images in mosaique (some images are 10000x5000)

rpautrat commented 1 year ago

The large resolution of images might indeed be an issue. DeepLSD is usually working on images of side 500-1000 px. For larger resolutions, the lines will become much more fragmented and noisy. But if each sub-image in your mosaique is smaller than 1000 px, the default parameters should be fine I think. The released model has been trained with 512x512 images for example.