On the Performance of GoogLeNet and AlexNet Applied to Sketches

Authors: Pedro Ballester, Ricardo Araujo

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our general methodology consists of using CNN trained over the Image Net data set and applying the resulting model to the sketch data set, registering the classification for each example in each category and analyzing the results.
Researcher Affiliation Academia Pedro Ballester and Ricardo Matsumura Araujo Center for Technological Development, Federal University of Pelotas Pelotas, RS, Brazil pedballester@gmail.com, ricardo@inf.ufpel.edu.br
Pseudocode No The paper describes the methods and evaluation in prose but does not include any structured pseudocode or algorithm blocks.
Open Source Code No The paper mentions using the Caffe framework for implementation but does not state that the authors are releasing their own source code for the described methodology or provide a link.
Open Datasets Yes Our general methodology consists of using CNN trained over the Image Net data set and applying the resulting model to the sketch data set...Image Net is a large collection of hierarchical labeled images that is used in the Image Net Challenge (Russakovsky et al. 2015)...We use the TU-Berlin sketch (Eitz, Hays, and Alexa 2012) data set
Dataset Splits No The paper states that pre-trained models (Goog Le Net and Alex Net) were used on the Image Net dataset and then applied to the TU-Berlin sketch dataset for testing. It does not describe specific training, validation, and test splits for its own experimental setup, as it uses pre-trained models rather than training new ones.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., GPU models, CPU types, memory) used to run the experiments.
Software Dependencies No The paper mentions using 'Caffe (Jia et al. 2014), a Deep Learning framework' but does not specify any version numbers for Caffe or any other software dependencies.
Experiment Setup Yes Our general methodology consists of using CNN trained over the Image Net data set and applying the resulting model to the sketch data set, registering the classification for each example in each category and analyzing the results. Both CNN output a probability distribution over possible classes for the input. Two different methods were used to evaluate the results. The first one considers only the top 10 most probable classes and the second one register the position of the correct class in the full probability ranking. We calculate the mean and standard deviation for each category.