JHLee0513 / semantic_bevnet

Code repository for Semantic Terrain Classification for Off-Road Autonomous Driving (https://openreview.net/forum?id=AL4FPs84YdQ) (CoRL 2021)
58 stars 12 forks source link

BEVNet

Source code for our work "Semantic Terrain Classification for Off-Road Autonomous Driving"

TODOs

Datasets

Datasets should be put inside data/. For example, data/semantic_kitti_4class_100x100.

Download links

Running the pretrained models

Model weights

SemanticKITTI

RELLIS

Run the models

First extract the model weights

cd /path/to/bevnet/experiments
unzip /path/to/zip/file

To run the models on the validation set, cd to bevnet/bevnet, then run

# Single-frame model
python test_single.py --model_file ../experiments/kitti4_100/single/include_unknown/default-logs/model.pth.4 --test_env kitti4

# Recurrent model
python test_recurrent.py --model_file ../experiments/kitti4_100/recurrent/include_unknown/default-logs/model.pth.2 --test_env kitti4

Training

BEVNet-S

Example:

cd experiments
bash train_kitti4-unknown_single.sh kitti4_100/single/include_unknown/default.yaml <tag> arg1 arg2 ...

Logs and model weights will be stored in a subdirectory of the config file like this: experiments/kitti4_100/single/include_unknown/default-<tag>-logs/

BEVNet-R

The command line formats are the same as BEVNet-S Example:

cd experiments
bash train_kitti4-unknown_recurrent.sh kitti4_100/recurrent/include_unknown/default.yaml <tag> \
--n_frame=6 --seq_len=20 --frame_strides 1 10 20 \
--resume kitti4_100/single/include_unknown/default-logs/model.pth.4 \
--resume_epoch 0

Logs and model weights will be stored in a subdirectory of the config file experiments/kitti4_100/recurrent/include_unknown/default-<tag>-logs/.