Label Encoding for Regression Networks

Authors: Deval Shah, Zi Yu Xue, Tor Aamodt

ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate BEL on four complex regression problems: head pose estimation, facial landmark detection, age estimation, and end-to-end autonomous driving.
Researcher Affiliation Academia Deval Shah, Zi Yu Xue & Tor M. Aamodt Department of Electrical and Computer Engineering University of British Columbia, Vancouver, BC, Canada
Pseudocode No The paper describes methods through text and figures (e.g., Figure 1 for training/inference flow), but does not contain a dedicated pseudocode block or algorithm listing.
Open Source Code Yes Code is available at https://github.com/ubc-aamodt-group/BEL_regression. We have provided the training and inference code with trained models.
Open Datasets Yes We follow the evaluation setting of Hopenet (Ruiz et al., 2018) and FSA-Net (fsa, 2019) and use two evaluation protocols with three widely used datasets: 300W-LP (Zhu et al., 2016), BIWI (Fanelli et al., 2013), and AFLW2000 (Zhu et al., 2016).
Dataset Splits Yes In these experiments 20% of the training set is used as validation set and the validation error is used to choose the best BEL approach.
Hardware Specification Yes All experiments are conducted on a Linux machine with an Intel i9-9900X processor and an Nvidia RTX 2080 Ti GPU with 11GB of memory.
Software Dependencies Yes Our code is implemented using Python 3.8.3 with Pytorch 1.5.1 using CUDA 10.2.
Experiment Setup Yes We use two runs with different random seeds for each combination of learning rate {0.001, 0.0001, 0.00001} and batch size {8, 16} are used for hyperparameter tuning.