Closed alighofrani95 closed 3 years ago
Where did you obtain the camera poses? From the software that comes with the T265?
If the poses are not accurate, you can run the reconstruction from scratch using pipeline_SfM.ipynb
. Otherwise, run the triangulation as in pipeline_Aachen.ipynb
. When matching the database images, exhaustive matching might be too slow so maybe use sequential matching instead - you'll need to write yourself the script that generates the images pairs. For the query, I recommend image retrieval using pairs_from_retrieval.py
and a global descriptor like DIR or OpenIBL.
Closing due to inactivity. Feel free to reopen if needed.
Hi @Skydes . Thank you for your thorough explanations. Forgive me for my lack of understanding w.r.t the topic, but it is possible to use NetVLAD as the global descriptor apart from the aforementioned descriptors (DIR and OpenIBL), right?
Thanks for your time.
Hi Sarlin,
I'm so exciting this implementation and want to use this approach on my own dataset, as I follow the Aachen dataset
I start to capture hundred of images from my room by Intel T265 camera and create dataset that consist of imgname#numer.jpg X Y Z w p q r (exports file_name, cartesian position, and quaternion values) Dose it need to save any other information?
Could you please give me step by step guide to use SuperGlue+SuperPoint to predict the position of a new image from my place?