Closed yilinliu77 closed 5 years ago
The estimate_transform-mesh is only able to align already roughly aligned meshes, it's an implementation of iterative closest point (ICP) algorithm. If the meshes are not roughly aligned beforehand it will converge to a wrong local minima, but does this deterministically so multiple invocations should not change the result.
How well is the alignment of the meshes before the invocation of the app?
I got it. I import the flight data into my reconstruction tools and it works. The reconstructed model is belong the input data's coordinates. Thanks for your explanation!
Hello, I have reconstructed the mesh model just like you. Now I want to compare my mesh with ground truth, but I can't find the scene binary file (e.g. NY-1.bin) anywhere. Could you tell me where the file is?
Looking forward to your reply. Thank you.
I found the Evaluation Tools in the website. After reading the README, I use the trajectory provided in NY-1 and generate the image successfully. Then I import the images and reconstruct a model using Reality Capture .
After that, I want to use the evaluation tools to evaluate the model. So I can evaluate my own trajectory. Here is my workflow of the reconstruct model:
After that I found the evaluation scores are very low, only 0.00267581 completeness when 0.0075. So i'm wondering if the
convert
is not work. Because no matter how many times I use theconvert
tools, it's average distance is the same.(0.09 for the gesox-mesh_orig.ply, and 0.12 for the model reconstructed by the images provided in the dataset).Did I misuse the tools?