search

ducha-aiki / navigation-benchmark

Code for "Benchmarking Classic and Learned Navigation in Complex 3D Environments" paper
63 stars 11 forks source link

Fine-tuning code for MP3D monodepth? #1

Open mmaz opened 5 years ago

mmaz commented 5 years ago

Hi, I am using Habitat and read in your paper/code that this project has made some pretrained weights available in habitat_baselines/agents/slam_agents.py for the ResNet50 variant of this depth estimation network by Hu et al, finetuned for Matterport3D scenes. Would it be possible to provide a pointer to the fine-tuning training code for the Matterport3D weights? I would like to see which scenes were used to fine-tune (maybe all scenes?) and try finetuning while holding some MP3D scenes out (as part of validation in my custom Habitat task).

Also, thank you so much for providing all these great embodied agent navigation tools (including the python ORBSLAM2 bindings!)

ducha-aiki commented 5 years ago

Hi,

Thanks for interest in out work. The finetuning was done by @dosovits , it is better to ask him about it :)