facebookresearch / OrienterNet

Source Code for Paper "OrienterNet Visual Localization in 2D Public Maps with Neural Matching"
Other
463 stars 48 forks source link

How to switch from "single-frame mode" to "sequence mode" in this project? #10

Open carrroo opened 1 year ago

carrroo commented 1 year ago

Hello, the paper pointed out that the accuracy of the image sequence is higher, so how to switch from "single-frame mode" to "sequence mode" in this project? :) Thanks!

carrroo commented 1 year ago

I tried to replace the file name in the "test2_files.txt" with a sequence of images, however the accuracy didn't seem to improve

sarlinpe commented 1 year ago

From the README:

To run the evaluation in sequential mode (by default with 10 frames): python -m maploc.evaluation.mapillary [...] --sequential

maploc.evaluation.kitti has the same flag. The maximum number of frames per segment can be updated with:

python -m maploc.evaluation.mapillary [...] --sequential chunking.max_length=20

I have updated the README accordingly.

carrroo commented 1 year ago

I saw the parameter configuration in "kitti.py" just recently, appreciate your reply

carrroo commented 1 year ago

_File "/home/zj/OrienterNet/maploc/data/sequential.py", line 18, in chunk_sequence sort_array = data.get("capturetime", data.get("index", names or indices)) ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

After I add the "--sequential" flag, it still doesn't seem to work. Is sequential mode not suitable for KITTI?

sarlinpe commented 1 year ago

Pushed a fix in https://github.com/facebookresearch/OrienterNet/commit/213aff45ce49a6aea11d273d198d9c2969457e10.

carrroo commented 1 year ago

Thank you very much for your reply! I noticed that the function _plot_examplesequential is not implemented, will sequential mode support visual export in the future?

sarlinpe commented 1 year ago

Yes, this is WIP in this notebook. I'll add code to generate sequential animations when I find the time.