Skip to content

Model pairs

tforgaard requested to merge model_pairs into main

  • create_img_list creates img lists from transitions_cropped

  • Have made some partial pipelines which allow for more efficient batch jobbing (hopefully)

    • pipeline_features first extract global and local features for all video ids (Needs GPU)
    • pipeline_matching then runs pairing and matching for all video ids, all parts, (multiple parts in a video!) and sets up everything for running reconstructions (Needs GPU)
    • pipeline_reconstruction then runs reconstructions with pycolmap for all video ids, all parts. (Only needs CPU I think)

  • Find_model_pairs creates a json file with scores for model pairings

TODO: @skalanan create q partial pipeline which load the json file and starts with the pairs that have the highest scores and then run localization pipeline. NB! Find_model_pairs already does retrieval/pairing, no need to do it again in the localization pipeline, just check if the text files exist and then just to matching

Merge request reports