SOD3D: A salient object detection dataset for photogrammetric 3D reconstruction
Description
SOD3D is a testing dataset built to evaluate the impact of an SOD stage applied to 3D reconstruction. It consists of images of 28 objects of different sizes, shapes and textures, and their processed versions. Starting with 36 original photographs taken for every object from various points of view, a manually segmented ground truth is included for every original image as a baseline for SOD algorithm evaluation. Furthermore, saliency maps generated with four SOD algorithms, one symbolic, based on the brain programming methodology, and three sub-symbolic, based on DL techniques, are included. Additionally, binarized versions of every saliency map are included to evaluate the performance of the SOD algorithms. Finally, we include a set of image overlays masking the original images with the proto-objects. The repository’s root contains a directory for every object. Inside these are a folder for the original images and their processed versions. Inside every object’s folder, there is a total of six folders that contain the original images, manually segmented ground truth, saliency maps generated by every SOD algorithm, the proto-objects that derive from the saliency maps, the overlays generated from masking the original images with the proto-objects and the final reconstructions in the form of point clouds resulting for every object after applying every proposed technique. The Original and GroundTruth folders contain the raw images and their manually segmented versions. Then, the SaliencyMap and ProtoObject folders share the same structure, where an independent directory is reserved for every SOD algorithm. Similarly, Overlay includes a folder for the data derived from every SOD algorithm, with the addition of a folder called GroundTruth, which contains the data derived from processing the original images using the manually segmented ground truth, to serve as a reference. Finally, the Reconstruction folder contains the available 3D reconstructions in the form of poing clouds for every applied technique.
Files
Steps to reproduce
This repository contains 28 sets of high-resolution images of different everyday objects. Each set comprises 36 images of every object, acquired from various camera positions with the primary goal of reconstructing the object through photogrammetric techniques. The 1,008 original images were taken with a Canon EOS 5D Mark III camera and a 24–105 mm variable lens, keeping optical parameters as fixed as possible to standardize image acquisition. To evaluate salient object detection (SOD) algorithms, the database includes manually segmented ground truth for each original image (Adobe Photoshop CS6 13.0). Furthermore, the repository contains automatically segmented versions of the images processed with three state-of-the-art deep learning (DL) based SOD algorithms (Python 3.10.16) and an additional set of images processed with a symbolic SOD algorithm based on the brain programming paradigm (Matlab R2017b). Finally, for 3D reconstructions, the database includes the 3D models for each original object (Meshroom-2021.1.0 [1]).
Institutions
- Centro de Investigacion Cientifica y de Educacion Superior de EnsenadaBaja California, Ensenada