Counterexample-Guided Data Augmentation

Authors: Tommaso Dreossi, Shromona Ghosh, Xiangyu Yue, Kurt Keutzer, Alberto Sangiovanni-Vincentelli, Sanjit A. Seshia

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

Reproducibility Variable Result LLM Response
Research Type Experimental We show the efficacy of the proposed framework by comparing it to classical augmentation techniques on a case study of object detection in autonomous driving based on deep neural networks.
Researcher Affiliation Academia Tommaso Dreossi, Shromona Ghosh, Xiangyu Yue, Kurt Keutzer, Alberto Sangiovanni-Vincentelli and Sanjit A. Seshia University of California, Berkeley {dreossi, shromona.ghosh, xyyue}@berkeley.edu, {keutzer, alberto, sseshia}@eecs.berkeley.edu
Pseudocode No The paper describes the augmentation scheme steps in paragraph text and presents a diagram (Figure 1), but does not include structured pseudocode or algorithm blocks.
Open Source Code Yes The implementation of the proposed framework and the reported experiments are available at https: //github.com/dreossi/analyze NN.
Open Datasets No In our experiments, we use synthetic data sets generated by our image generator.
Dataset Splits No The original training and test sets X and T contain 1500 and 750 pictures, respectively, randomly generated by our image generator.
Hardware Specification Yes We acknowledge the support of NVIDIA Corporation via the donation of the Titan Xp GPU used for this research.
Software Dependencies No The paper mentions using 'squeeze Det' and 'imgaug' (with a footnote link to its GitHub page: https://github.com/aleju/imgaug) but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes All models were trained for 65 epochs.