Neighbourhood Consensus Networks

Authors: Ignacio Rocco, Mircea Cimpoi, Relja Arandjelović, Akihiko Torii, Tomas Pajdla, Josef Sivic

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The proposed approach was evaluated on both instance- and category-level matching problems. The same approach is used to obtain reliable matches for both problems, which are then used to solve two completely different tasks: (i) camera pose estimation in the challenging scenario of indoor localization, in the instance-level matching case, and (ii) semantic object alignment in the category-level matching case.
Researcher Affiliation Collaboration Ignacio Rocco Mircea Cimpoi Relja Arandjelovi c Akihiko Torii Tomas Pajdla Josef Sivic , Inria CIIRC, CTU in Prague Deep Mind Tokyo Institute of Technology
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks (clearly labeled algorithm sections or code-like formatted procedures).
Open Source Code Yes Both training code and models are available at [1]. [1] Project webpage (code/networks). http://www.di.ens.fr/willow/research/ncnet/.
Open Datasets Yes We report results on the PF-Pascal [11] dataset, which contains 1,351 semantically related image pairs from the 20 object categories of the PASCAL VOC [9] dataset. ... We evaluate on the In Loc dataset [42]... As the In Loc dataset was designed for evaluation and does not contain a training set, we collected an Indoor Venues Dataset, consisting of user-uploaded photos... The design and collection procedures are described in the supplementary material [30] and the dataset is available at [1].
Dataset Splits Yes We follow the same evaluation protocol as [12, 29], and use the split from [12] which divides the dataset into approximately 700 pairs for training, 300 for validation and 300 for testing. ... This dataset contains 3861 positive image pairs from 89 different venues in 6 different cities, split into train: 3481 pairs (80 places) and validation: 380 pairs (from the remaining 9 places).
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions 'implemented in Py Torch' but does not provide specific version numbers for PyTorch or any other software dependencies.
Experiment Setup Yes The model is initially trained for 5 epochs using Adam optimizer [20], with a learning rate of 5 10 4 and keeping the feature extraction layer weights fixed. For category level matching, the model is then subsequently finetuned for 5 more epochs, training both the feature extraction and the neighbourhood consensus network, with a learning rate of 1 10 5.