PPT:
- Full point cloud - car suraface at the origin
- show reconstruction
- next step - put points at a transform
- goal is to optimize even now, estimate transform
- show visualization - mention need for good initialization
./install.sh
python ours/sliding_window_reconstruct_multiple_frame_argo.py --config configs/config_kitti.json --sequence_dir data/P04/cleaned_data/000/000003/pcd.npy
python dsp_slam/reconstruct_frame.py --config configs/config_kitti.json --sequence_dir data/dsp_slam/kitti/07 --frame_id 0
python dsp_slam/reconstruct_frame.py --config configs/config_kitti.json --sequence_dir data/dsp_slam/kitti/07 --frame_id 50
python dsp_slam/reconstruct_frame.py --config configs/config_kitti.json --sequence_dir data/dsp_slam/kitti/07 --frame_id 100
python dsp_slam/reconstruct_frame.py --config configs/config_kitti.json --sequence_dir data/dsp_slam/kitti/07 --frame_id 500
python dsp_slam/reconstruct_frame.py --config configs/config_kitti.json --sequence_dir data/dsp_slam/kitti/07 --frame_id 1000
python dsp_slam/reconstruct_frame_no_render_term.py --config configs/config_kitti.json --sequence_dir data/dsp_slam/kitti/07 --frame_id 0
python dsp_slam/reconstruct_frame_no_render_term.py --config configs/config_kitti.json --sequence_dir data/dsp_slam/kitti/07 --frame_id 50
python dsp_slam/reconstruct_frame_no_render_term.py --config configs/config_kitti.json --sequence_dir data/dsp_slam/kitti/07 --frame_id 100
python dsp_slam/reconstruct_frame_no_render_term.py --config configs/config_kitti.json --sequence_dir data/dsp_slam/kitti/07 --frame_id 500
python dsp_slam/reconstruct_frame_no_render_term.py --config configs/config_kitti.json --sequence_dir data/dsp_slam/kitti/07 --frame_id 1000
- Run all with table.
python3 ours/run_all.py
- Run all bash no table
./run_all.sh
- Evaluate optimization
python ours/2_reconstruct_multiple_frame_argo_compare.py --config configs/config_kitti.json --sequence_dir data/P04/cleaned_data/000/000009/pcd.npy
python ours/2_reconstruct_multiple_frame_argo_compare.py --config configs/config_kitti.json --sequence_dir data/P04/cleaned_data/002/002001/pcd.npy
# motion
# Pose esimation/ nosiy
# adn reconstruction chamfer distance
# inconsidtent over time and no using for data of mulitple frame
# ablation first frame vs multiple frame
# methodology tracking optimization, kiss icp pv-rcnn deepsdf
# abstraction -> Briftly introduction 200 words
# 1. introduction: why? what? how? *Claims
# 2. Related works: mention the exising works
# 3. methodology: how we do that this
# 4. Experiemtn: explain result, desscribe dataset, evaluatuoin metric, baseline(yue pan paper), implementation deitail(parameter setting, learnng, ) show result. abltion, change some params -> should support claim
# 5. Conclusion: repeat abstraction and disadvatange , futre work how can we imrpove
Shashank:
- The naming of pts_cam_pcd, T_cam_obj seems to be wrong. velodyne in place of cam seems correct to me.
- [NOT TRUE] The surface_points and pts_cam_pcd are not with a relative transform of T_cam_obj. small difference. Visible with car_no = 1; frame = 10