NCNet Re-implementation on Pytorch 1.0 + Python3.6 (Original implementation : https://github.com/ignacio-rocco/ncnet) For more information check out their project website and their paper on arXiv.
More visual results can be found here
| TOP 5 % Matches | TOP 10 % Matches | TOP 20 % Matches | Ground Truth | Key Points Match |
|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
To download Pre-trained feature extractor (ResNet 18 ):
cd model/FeatureExtractor
bash download.shTo download PF Pascal Dataset :
cd data/pf-pascal/
bash download.shFor important functions, we provide a quick search here
To train on PF-Pascal :
bash demo_train.shTo evaluate on PF-Pascal :
bash demo_eval.sh