Description (include details on usage, files and paper references) | This dataset provides a collection of web images and 3D models for research on landmark recognition (especially for methods based on 3D models). We hope it could serve as a useful benchmark for evaluating and comparing different methods.
Contents of the Dataset
Currently the dataset involves:
25 landmarks, each with a 3D model (point cloud) reconstructed using structure-from-motion (SfM);
45,180 database images (1.4K~2K per landmark) registered to the 3D models;
10,000 positive test images for evaluation (400 per landmark);
~2.7 million 3D points in the 3D models (29K~223K per landmark);
~58 million SIFT features registered to the 3D models.
The list of landmarks is provided in the file list_landmarks.txt in the root folder.
Qiang Hao, Rui Cai, Zhiwei Li, Lei Zhang, Yanwei Pang, and Feng Wu. "3D Visual Phrases for Landmark Recognition". in Proc. of the 25th IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2012.
For each landmark, there is a subfolder containing four files:
list_db.txt -- the list of database images
list_test.txt -- the list of test images
model.nvm -- the 3D model reconstructed using VisualSFM
model.ply -- the 3D points in the 3D model (for visualization only) |
|