Skip to content

Latest commit

 

History

History
90 lines (73 loc) · 3.18 KB

adaptor.md

File metadata and controls

90 lines (73 loc) · 3.18 KB
Raw

Merging Map Data with Trajectory Data

Necessary Files

  • mapping_results.pickle from mapping evaluation
  • bev_features stored in previous /path_to_save_bev_features
  • Trajdata pickle files: traj_scene_frame_{full_train, full_val, mini_train, mini_val}.pkl
  • Ground truth files: gt_{full_train, full_val, mini_val}.pickle

You can download the above Trajdata pickle files and Ground truth files files from here

Data Merging Process

First create a new directory to store the above files, and a new directory to store the merged map & trajectory data

mkdir adaptor_files trj_data
cd trj_data
mkdir maptr maptrv2 maptrv2_cent stream
cd ..

Folder Structure

MapBEVPrediction
├── nuscenes/
├── processed/
├── adaptor_files/
│   ├── traj_scene_frame_{full_train, full_val, mini_train, mini_val}.pkl
│   ├── gt_{full_train, full_val, mini_val}.pickle
│   ├── mapping_results.pickle
├── trj_data/
│   ├── maptr/
│   ├── maptrv2/
│   ├── maptrv2_cent/
│   ├── stream/

Merging Map Data and Trajectory Data

Run the script adaptor.py in MapBEVPrediction/adaptor/ to merge mapping data and trajectory data.

cd adaptor

# To merge MapTRv2 centerline, add in an additional argument --centerline
python adaptor.py \
  --version mini \                                       # [trainval, mini]
  --split mini_val \                                     # [train, train_val, val, mini_train, mini_val]
  --map_model MapTR \                                    # [MapTR, StreamMapNet]
  --dataroot ../nuscenes \
  --index_file ../adaptor_files/traj_scene_frame_mini_val.pkl \
  --map_file ../adaptor_files/mapping_results.pickle \
  --gt_map_file ../adaptor_files/gt_mini_val.pickle \
  --save_path ../trj_data/maptr                          # trj_data directory, [maptr, maptrv2, maptrv2_cent, stream]

# For visualization
python adaptor_vis.py --data_path ../trj_data/maptr

After running the above script, folder structure should look like this:

MapUncertaintyPrediction
├── nuscenes/
├── processed/
├── adaptor_files/
├── trj_data/
|   ├── maptr/
│   |   ├── mini_val/
│   |   |   ├── data/
│   |   |   |   ├── scene-{scene_id}.pkl
│   |   ├── train/
│   |   ├── val/
│   ├── maptrv2/
│   ├── maptrv2_cent/
│   ├── stream/

Merging in BEV Features

To merge in the intermediate BEV features produced during map estimation, run the following:

python adaptor_bev.py \
  --scene_folder /path_to_scene_folder \
  --bev_folder /path_to_save_bev_features \
  --map_model {MapTR, StreamMapNet}

Note that this merging process needs to be conducted at least twice per mapping method. Once on train split, once on val split. You need mapping results from evaluating on both the train and val set for each mapping method. So in total, the mapping evaluation scripts should be run 8 times (2 times 4 mapping methods). If you only want to examine one of the mapping method, then 2 times are enough.