Skip to content

Latest commit

 

History

History
84 lines (53 loc) · 3.81 KB

File metadata and controls

84 lines (53 loc) · 3.81 KB

Deposit dataset

kitti

** branch: https://github.com/eric-yyjau/kitti_instance_RGBD_utils/tree/master/kitti_tools **

In kitti_tools:

python dump_img_odo_tum.py --dump --dataset_dir /data/kitti/odometry --with_pose --with_X --with_sift --dump_root /newfoundland/yyjau/kitti/kitti_dump/odo_corr_dump_siftIdx_npy_delta1235810_full_1027 --num_threads=1 --img_height 376 --img_width 1241 --dataloader_name kitti_seq_loader --cam_id '02'

apollo

** branch: https://github.com/eric-yyjau/kitti_instance_RGBD_utils/tree/dump_data/kitti_tools**

  • Download dataset:

  • Process data

No need to process data. Already matched.

  • Dump data: (make sure to specify dimensions)

In kitti_tools:

python dump_img_odo_tum.py --dump --dataset_dir /newfoundland/yyjau/apollo/train_seq_1/ --dataloader_name apollo_train_loader --with_pose --with_sift --dump_root /newfoundland/yyjau/apollo/apollo_dump/train_seq_1/ --num_threads=1 --cam_id 5 --img_height 2710 --img_width 3384

  • deprecated data loader (for self_localization_examples.tar.gz. Images not good.)

python dump_img_odo_tum.py --dump --dataset_dir /newfoundland/yyjau/apollo/sample_1/ --dataloader_name apollo_seq_loader --with_pose --with_sift --dump_root /newfoundland/yyjau/apollo/apollo_dump/sample_3/ --num_threads=1 --cam_id 1 --img_height 480 --img_width 600

tum

** branch: https://github.com/eric-yyjau/kitti_instance_RGBD_utils/tree/dump_data/kitti_tools**

  • Download dataset: (run the script in the folder to download)

python kitti_tools/tum/download.py

  • Match time stamps of rgb images, depth images and poses

In kitti_tools:

python process_poses.py --dataset_dir [path to dataset]

  • Dump data: (make sure to specify dimensions)

In kitti_tools:

python dump_img_odo_tum.py --dump --dataset_dir /data/tum/raw_sequences --with_pose --with_sift --dump_root /data/tum/tum_dump/slam_seq_v1 --with_X --num_threads=1 --cam_id 00 --img_height 480 --img_width 640 --dataloader_name tum_seq_loader

Euroc

** branch: https://github.com/eric-yyjau/kitti_instance_RGBD_utils/tree/dump_data/kitti_tools**

  • Download dataset: (run the script in the folder to download)

python kitti_tools/euroc/download.py

  • Process data - Match time stamps of rgb images, depth images and poses

In kitti_tools:

python process_poses_euroc.py --dataset_dir /data/euroc/test2/ --dataset euroc

  • Dump data: (make sure to specify dimensions)

In kitti_tools:

python dump_img_odo_tum.py --dump --dataset_dir /data/euroc/raw_sequence --with_pose --with_sift --dump_root /data/euroc/euroc_dump/slam_seq_v1 --num_threads=1 --cam_id 00 --img_height 480 --img_width 752 --dataloader_name euroc_seq_loader

Run training

Env:

conda create -n kitti_py36 python=3.6 pip conda activate kitti_py36 Install PyTorch. pip install -r requirements.txt export SHAPER_MODELS_PATH='/home/ruizhu/Documents/Projects/kitti_instance_RGBD_utils/deepSfm_ori/models/shaper/shaper/models'

Training command:

CUDA_VISIBLE_DEVICES=1 python train_good_corr_4_vals_goodF_baseline.py train_good configs/kitti_corr_baseline.yaml temp --evalCUDA_VISIBLE_DEVICES=1 python train_good_corr_4_vals_goodF_baseline.py train_good configs/kitti_corr_baseline.yaml temp --eval

SuperPoint input

change if_SP to true in .config file to enable input from SP instead of SIFT.