Unsupervised Object Segmentation by Redrawing

Authors: Mickaël Chen, Thierry Artières, Ludovic Denoyer

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experiment with this method on different datasets and demonstrate the good quality of extracted masks. (...) Finally, we present experimental results on three datasets in Section 5 that explore the feasibility of unsupervised segmentation within our framework and compare its performance against a baseline supervised with few labeled examples.
Researcher Affiliation Collaboration Mickaël Chen Sorbonne Université, CNRS, LIP6, F-75005, Paris, France mickael.chen@lip6.fr Thierry Artières Aix Marseille Univ, Université de Toulon, CNRS, LIS, Marseille, France Ecole Centrale Marseille thierry.artieres@centrale-marseille.fr Ludovic Denoyer Facebook Artificial Intelligence Research denoyer@fb.com
Pseudocode Yes The final learning algorithm follows classical GAN schema [3, 17, 39, 57] by updating the generator and the discriminator alternatively with the update functions presented in Algorithm 1. (...) Algorithm 1 Networks update functions
Open Source Code Yes The code, dataset splits and pre-trained models are also available open-source 1. 1https://github.com/mickael Chen/Re DO
Open Datasets Yes Flowers dataset [40, 41] is composed of 8189 images of flowers. (...) Labeled Faces in the Wild dataset [25, 33] is a dataset of 13233 faces. (...) The Caltech-UCSD Birds 200 2011 (CUB-200-2011) dataset [53] is a dataset containing 11788 photographs of birds. (...) colored MNIST [34] numbers:
Dataset Splits Yes Flowers dataset [40, 41] (...) We split into sets of 6149 training images, 1020 validation and 1020 test images (...) Labeled Faces in the Wild dataset [25, 33] (...) The test set is composed of 1600 images, and the validation set of 1327 images. (...) The Caltech-UCSD Birds 200 2011 (CUB-200-2011) dataset [53] (...) We use 10000 images for our training split, 1000 for the test split, and the rest for validation.
Hardware Specification Yes We used mini-batches of size 25 and ran each experiment on a single NVidia Tesla P100 GPU.
Software Dependencies No The paper mentions using ADAM optimizer and inspirations from PSPNet and SAGAN architectures, but does not provide specific version numbers for software dependencies or libraries like Python, PyTorch, or TensorFlow.
Experiment Setup Yes Learning rates are set to 10 4 except for the mask network F which uses a smaller value of 10 5. We sample noise vectors zi of size 32 (except for MNIST where we used vectors of size 16) from N(0, Id) distribution. We used mini-batches of size 25 and ran each experiment on a single NVidia Tesla P100 GPU.