Learning feed-forward one-shot learners

Authors: Luca Bertinetto, João F. Henriques, Jack Valmadre, Philip Torr, Andrea Vedaldi

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate learnets against baseline one-shot architectures (sect. 3.1) on two one-shot learning problems in Optical Character Recognition (OCR; sect. 3.2) and visual object tracking (sect. 3.3). All experiments were performed using Mat Conv Net [22].
Researcher Affiliation Academia Luca Bertinetto University of Oxford luca@robots.ox.ac.uk, João F. Henriques University of Oxford joao@robots.ox.ac.uk, Jack Valmadre University of Oxford jvlmdr@robots.ox.ac.uk, Philip H. S. Torr University of Oxford philip.torr@eng.ox.ac.uk, Andrea Vedaldi University of Oxford vedaldi@robots.ox.ac.uk
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any concrete access to source code for the methodology described.
Open Datasets Yes For this, we use the Omniglot dataset [13], which contains images of handwritten characters from 50 different alphabets.
Dataset Splits Yes In this expression, the performance of the predictor extracted by the learnet from the exemplar zi is assessed on a single validation pair (xi, ℓi)
Hardware Specification No The paper mentions 'on a GPU' when discussing a related work (SO-DLT), but does not provide specific hardware details (like GPU models, CPU types, or memory) used for its own experiments.
Software Dependencies No The paper states 'All experiments were performed using Mat Conv Net [22]', but does not provide a specific version number for this software or any other dependencies.
Experiment Setup Yes The baseline stream ϕ for the siamese, siamese learnet, and single-stream learnet architecture consists of 3 convolutional layers, with 2 2 max-pooling layers of stride 2 between them. The filter sizes are 5 5 1 16, 5 5 16 64 and 4 4 64 512. ... All networks are trained from scratch using SGD for 50 epoch of 50,000 sample triplets (zi, xi, ℓi). The same hyperparameters (learning rate of 10 2 geometrically decaying to 10 5, weight decay of 0.005, and small mini-batches of size 8) are used for all experiments...