EvidMTL: Evidential Multi-Task Learning for Uncertainty-Aware Semantic Surface Mapping from Monocular RGB Images
Paper โข 2503.04441 โข Published
This repository hosts the trained checkpoints for Evidential Mapping, kept separate from the source repository.
Paper: EvidMTL: Evidential Multi-Task Learning for Uncertainty-Aware Semantic Surface Mapping from Monocular RGB Images (IROS 2025, DOI: 10.1109/IROS60139.2025.11247528).
checkpoints/silog_klmuprior_alpha2_beta0.1_gamma1.0_no_noise_final_model.ckpt:
SiLog+KL model using the ground-truth mean directly as the prior. This is
the final checkpoint from W&B run 4slgriz2, trained from source commit
80e0f76567fcc708fd2c7fca9849fa554d954f53.checkpoints/silog_klmuprior_alpha1.01_gamma0.001_betacalc_final_model.ckpt:
KL model using a noisy ground-truth mean prior.The corresponding initial source release is based on EvidMTL branch wonka
at commit 7bd91ec6d6957c9cfee19db1b29aab81bb2d3162. See the
HumanoidsBonn/evidmtl repository
for environment, configuration, and evaluation details.
These EvidMTL artifacts are released under GPL-3.0, matching the EvidMTL source license.