search

MarSaKi / VLN-BEVBert

[ICCV 2023} Official repo of "BEVBert: Multimodal Map Pre-training for Language-guided Navigation"
184 stars 5 forks source link

R2R-CE pre-training #4

Closed yhl2018 closed 1 year ago

yhl2018 commented 1 year ago

执行一下代码一直卡在以下界面呢？

for R2R-CE pre-training

python precompute_features/grid_habitat_clip.py 1698676417297

python precompute_features/save_habitat_img.py --img_type depth

1698676433321

MarSaKi commented 1 year ago

I guess you should re-compile the Matterport3D simulator in the cpu-mode, because there is conflict between habitat_sim and mp3d_sim_gpu. You can run this command to compile mp3d_sim in CPU mode: ' mkdir build_cpu && cd build_cpu cmake -DOSMESA_RENDERING=ON .. make cd ../ '

yhl2018 commented 1 year ago

1698757564389 CLIP-ViT-B-16-views-habitat.hdf5 这个文件是哪个脚本生成的啊？没有找到对应的名字呢？

yhl2018 commented 1 year ago

@MarSaKi 麻烦问您一下，训练ETPnav 的pretrain，和现在的这个BEVBert代码是完全一样的吗？以下几个特征也都是x需要的吗

MarSaKi commented 1 year ago

The pre-training code for BEVBert and ETPNav is not exactly the same. You can download the “CLIP-ViT-B-16-views-habitat.hdf5“ file from here.