Skip to content

Latest commit

 

History

History

evalCorr

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 

Evaluation on sparse correspondences (MegaDepth and RobotCar )

Data

MegaDepth

We release our test MegaDepth dataset including 1600 pairs, can be downloaded here here.

Once it is downloaded, saved it into ../../data, the file structure should be :

./RANSAC-Flow/data/MegaDepth
├── MegaDepth_Train/
├── MegaDepth_Train_Org/
├── Val/
└── Test/

RobotCar

The RobotCar dataset can be downloaded from the Visual Localization Challenge (at the bottom of the site), or more precisely here.

This evaluation code takes the same structure as MegaDepth, which means you need to have a csv containing all the correspondences and the image paths. The file structure should be :

./RANSAC-Flow/data/RobotCar
├── img/
└── test6511.csv

The cvs file can be downloaded from here(~5G).

MOCO feature

MegaDepth, W/O Fine-tuning

Running :

python evaluation.py --outDir MOCO_MegaDepth_WO_FT MegaDepth

To get results with our fine alignment :

python getResults.py --coarsePth MOCO_MegaDepth_WO_FT_Coarse --finePth MOCO_MegaDepth_WO_FT_Fine --dataset MegaDepth --multiH MegaDepth

MegaDepth, W Fine-tuning

Running :

python evaluation.py --outDir MOCO_MegaDepth_FT --resumePth ../../model/pretrained/MegaDepth_TestFT.pth MegaDepth

To get results with our fine alignment :

python getResults.py --coarsePth MOCO_MegaDepth_FT_Coarse --finePth MOCO_MegaDepth_FT_Fine --dataset MegaDepth --multiH MegaDepth

RobotCar, W/O Fine-tuning

Running :

python evaluation.py --outDir MOCO_RobotCar_WO_FT RobotCar

To get results with our fine alignment :

python getResults.py --coarsePth MOCO_RobotCar_WO_FT_Coarse --finePth MOCO_RobotCar_WO_FT_Fine --multiH --dataset RobotCar RobotCar

RobotCar, With Fine-tuning

Running :

python evaluation.py --outDir MOCO_RobotCar_FT --resumePth ../../model/pretrained/RobotCar_TestFT.pth  RobotCar

To get results with our fine alignment :

python getResults.py --coarsePth MOCO_RobotCar_FT_Coarse --finePth MOCO_RobotCar_FT_Fine --multiH --dataset RobotCar RobotCar

ImageNet feature

Adding --imageNet when running evaluation.py with the above commands.