Zero-Shot Semantic Segmentation

Authors: Maxime Bucher, Tuan-Hung VU, Matthieu Cord, Patrick Pérez

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

Reproducibility Variable Result LLM Response
Research Type Experimental We report evaluations of ZS3Net on two datasets (Pascal-VOC and Pascal-Context) and in zero-shot setups with varying numbers of unseen classes. Compared to a ZSL baseline, our method delivers excellent performances, which are further boosted using self-training and semantic contextual cues.
Researcher Affiliation Collaboration Maxime Bucher valeo.ai maxime.bucher@valeo.com Tuan-Hung Vu valeo.ai tuan-hung.vu@valeo.com Matthieu Cord Sorbonne Université valeo.ai matthieu.cord@lip6.fr Patrick Pérez valeo.ai patrick.perez@valeo.com
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statements or links indicating that source code for the described methodology is publicly available.
Open Datasets Yes Experimental evaluation is done on the two datasets: Pascal-VOC 2012 [15] and Pascal Context [31]. Pascal-VOC contains 1, 464 training images with segmentation annotations of 20 object classes. Similar to [11], we adopt additional supervision from semantic boundary annotations [19] during training. Pascal-Context provides dense semantic segmentation annotations for Pascal-VOC 2010, which comprises 4, 998 training and 5, 105 validation images of 59 object/stuff classes.
Dataset Splits Yes Pascal-Context provides dense semantic segmentation annotations for Pascal-VOC 2010, which comprises 4, 998 training and 5, 105 validation images of 59 object/stuff classes.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, or memory) used for running its experiments.
Software Dependencies No The paper mentions software like Deep Labv3+, ResNet-101, SGD, Adam optimizer, and word2vec, but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes Segmentation models are trained by SGD [5] optimizer using polynomial learning rate decay with the base learning rate of 7e 3, weight decay 5e 4 and momentum 0.9. The GMMN is a multi-layer perceptron with one hidden layer, leaky-RELU non-linearity [29] and dropout [41]. In our experiments, we fix the number of hidden neurons to 256 and set the kernel bandwidths as {2, 5, 10, 20, 40, 60}. The input Gaussian noise has the same dimension as used w2c embeddings, namely 300. The generative model is trained using Adam optimizer [21] with the learning rate of 2e 4.