 NeurIPS'2020 submission, supplementary material.  
 Paper ID: 5984
 Paper title: "Soft Contrastive Learning for Visual Localization"
=====================================================================================================

This folder contains the following supplementary material:

1) A PDF file containing proofs as well as additional quantitative and visual results.

2) An example video. The video contains a comprehensive collection of Grad-CAM visualizations for three methods:

-Ours
-Off-the-shelf: Initialization with ImageNet without localization training
-Triplet trained on Oxford RobotCar

The query images in the video are linearly down-sampled to obtain a movie which fits the supplementary size requirements. For each query we show the query image, Grad-CAM on the query image, Grad-CAM on the retrieved image, the retrieved image, and the distance between retrieved image and query image. The color of the reported distance between query and reference image indicates whether an image was localized within the threshold for correct localization (10m).

3) Lists of
- all training images
- all validation query images
- all test reference images
- all test query images
Please note that, due to the 100Mb supplementary material size limit, the lists only include the image paths but not the image location and angle. After publication, we plan to make the full list, including coordinates, publicly available. 

