I actually do not want a pull request but wanted to post an issue.
I am trying to recreate a repo to evaluate 3D lidar data in kitti format.
I was able to train on model based on tyagi's point pillars, but as you mentioned it's flawed and can't evaluate(OOM or invalid tensor shape errors).
Can you help me? I am trying to run the same initialization commands through google colab but my setup fails. I needed to update CMakeLists.txt to not import the deleted file point_pillars_v2.cpp but even point_pillars.cpp hard setup errors. I was hoping perhaps you might have dealt with this before.
/content/PointPillars/src/point_pillars.cpp:348:7: error: redefinition of ‘float iou(const BoundingBox3D&, const BoundingBox3D&)’
float iou(const BoundingBox3D& box1,
^~~
/content/PointPillars/src/point_pillars.cpp:332:7: note: ‘float iou(const BoundingBox3D&, const BoundingBox3D&)’ previously defined here
float iou(const BoundingBox3D& box1,
^~~
/content/PointPillars/src/point_pillars.cpp: In function ‘float iou3D(const BoundingBox3D&, const BoundingBox3D&)’:
/content/PointPillars/src/point_pillars.cpp:374:18: error: ‘min’ was not declared in this scope
float topZ = min(box1.height/2 + box1.z, box2.height/2 + box2.z);
^~~
/content/PointPillars/src/point_pillars.cpp:374:18: note: suggested alternative:
In file included from /usr/include/c++/7/algorithm:62:0,
from /content/PointPillars/pybind11/include/pybind11/numpy.h:15,
from /content/PointPillars/src/point_pillars.cpp:2:
/usr/include/c++/7/bits/stl_algo.h:3456:5: note: ‘std::min’
min(initializer_list<_Tp> __l, _Compare __comp)
^~~
/content/PointPillars/src/point_pillars.cpp:375:21: error: ‘max’ was not declared in this scope
float bottomZ = max( - box1.height/2 + box1.z, - box2.height/2 + box2.z);
^~~
/content/PointPillars/src/point_pillars.cpp:375:21: note: suggested alternative:
Should I run setup.py through his repo and pull your other files in? would that work? Would training a model with his repo be the correct type of model for your evaluation/prediction?
I was hoping to be able to evaluate or even predict with 2D or 3D outputs.
Can you help me with this? Thank you for your time.
I tried second's point pillars but the preprocessing of my kitti files didn't generate a properly structured dict (pkl file) for their training or evaluation script and tyagi's ran smoothly. Visualization tips would greatly be appreciated.
I actually do not want a pull request but wanted to post an issue.
I am trying to recreate a repo to evaluate 3D lidar data in kitti format.
I was able to train on model based on tyagi's point pillars, but as you mentioned it's flawed and can't evaluate(OOM or invalid tensor shape errors).
Can you help me? I am trying to run the same initialization commands through google colab but my setup fails. I needed to update CMakeLists.txt to not import the deleted file point_pillars_v2.cpp but even point_pillars.cpp hard setup errors. I was hoping perhaps you might have dealt with this before.
/content/PointPillars/src/point_pillars.cpp:348:7: error: redefinition of ‘float iou(const BoundingBox3D&, const BoundingBox3D&)’ float iou(const BoundingBox3D& box1, ^~~ /content/PointPillars/src/point_pillars.cpp:332:7: note: ‘float iou(const BoundingBox3D&, const BoundingBox3D&)’ previously defined here float iou(const BoundingBox3D& box1, ^~~ /content/PointPillars/src/point_pillars.cpp: In function ‘float iou3D(const BoundingBox3D&, const BoundingBox3D&)’: /content/PointPillars/src/point_pillars.cpp:374:18: error: ‘min’ was not declared in this scope float topZ = min(box1.height/2 + box1.z, box2.height/2 + box2.z); ^~~ /content/PointPillars/src/point_pillars.cpp:374:18: note: suggested alternative: In file included from /usr/include/c++/7/algorithm:62:0, from /content/PointPillars/pybind11/include/pybind11/numpy.h:15, from /content/PointPillars/src/point_pillars.cpp:2: /usr/include/c++/7/bits/stl_algo.h:3456:5: note: ‘std::min’ min(initializer_list<_Tp> __l, _Compare __comp) ^~~ /content/PointPillars/src/point_pillars.cpp:375:21: error: ‘max’ was not declared in this scope float bottomZ = max( - box1.height/2 + box1.z, - box2.height/2 + box2.z); ^~~ /content/PointPillars/src/point_pillars.cpp:375:21: note: suggested alternative:
Should I run setup.py through his repo and pull your other files in? would that work? Would training a model with his repo be the correct type of model for your evaluation/prediction?
I was hoping to be able to evaluate or even predict with 2D or 3D outputs.
Can you help me with this? Thank you for your time. I tried second's point pillars but the preprocessing of my kitti files didn't generate a properly structured dict (pkl file) for their training or evaluation script and tyagi's ran smoothly. Visualization tips would greatly be appreciated.