High-dimensional Convolutional Networks for Geometric Pattern Recognition

Image Correspondences

Download YFCC100M Dataset

bash scripts/download_yfcc.sh /path/to/yfcc100m

Preprocess YFCC100M Dataset

SIFT

python -m scripts.gen_2d \
  --source /path/to/yfcc100m \
  --target /path/to/save/processed/dataset \

UCN

python -m scripts.gen_2d \ 
    --source /path/to/yfcc100m \
    --target /path/to/save/processed/dataset \
    --feature ucn \
    --onthefly \
    --ucn_weight /path/to/pretrained/ucn/weight

Training Network

Train an image correspondence network.

bash scripts/train_2d.sh "-experiment1" \
    "--data_dir_raw /path/to/raw/yfcc \
    --data_dir_processed /path/to/processed/yfcc"

Testing on YFCC100M Dataset

python -m scripts.benchmark_yfcc \
  --data_dir_raw /path/to/yfcc100m \
  --data_dir_processed /path/to/processed/dataset \
  --weights /path/to/checkpoint \
  --out_dir /path/to/save/outputs \
  --do_extract

Demo on YFCC100M Dataset

Following demo_2d script will download UCN and our best model(PyramidNetSCNoBlock) weights and test it on few pairs of images. The visualization output will be saved on './visualize' directory.

python demo_2d.py

Model Zoo

Model	Dataset	Link
PyramidNetSCNoBlock	YFCC100MDatasetUCN	download
ResNetSC	YFCC100MDatasetExtracted	download
ResUNetINBN2G	YFCC100MDatasetExtracted	download
OANet	YFCC100MDatasetExtracted	download
LFGCNet	YFCC100MDatasetExtracted	download

Raw data for Fig#

Prec-Recall

---