1. 09 Jul, 2020 1 commit
  2. 23 Jun, 2020 1 commit
  3. 15 Jun, 2020 1 commit
    • Jan Philipp Simon Langen's avatar
      change pipeline such that pca is immediately calculated when the max number of... · e7da2ad1
      Jan Philipp Simon Langen authored
      change pipeline such that pca is immediately calculated when the max number of features has been extracted. New features will instantly be scaled down to number of pca dims. This is to save memory. 5000 imgs of oxford with 1000 descriptors per image each containing 1024 dims saved as 32 byte floats would amount to ~164 GB of memory required. + Calculating PCA Matrix on this would require much more.
      
      Also now standardizing features (0 mean, 1 std) before calculating/applying pca. This is identical to calculating PCA on correlation instead of covariance matrix.
      e7da2ad1
  4. 13 Jun, 2020 1 commit
  5. 12 Jun, 2020 1 commit
  6. 11 Jun, 2020 1 commit
  7. 29 May, 2020 1 commit
  8. 25 May, 2020 1 commit
  9. 22 May, 2020 2 commits
  10. 08 May, 2020 1 commit
  11. 06 May, 2020 1 commit
  12. 04 May, 2020 1 commit
  13. 02 May, 2020 1 commit
  14. 01 May, 2020 1 commit
    • Jan Philipp Simon Langen's avatar
      - extract features now returns receptive field centers · 80b5805c
      Jan Philipp Simon Langen authored
      - wrote matching function (currently ransac runs on cpu single core - kinda slow, kd tree is multithreaded and fast)
      - matching works, tested on one of the query images, returns itself first, then the query images that show similar content then a few others rest is rejected
      
      Todos: matching parameters (thresholds, trails etc. need estimation)
      clean up
      incorperate into experiment structure
      process output information
      80b5805c
  15. 23 Apr, 2020 1 commit
    • Jan Philipp Simon Langen's avatar
      oxford_sorter.py is experimental. Mostly used it to get the index set but have... · 97df86b4
      Jan Philipp Simon Langen authored
      oxford_sorter.py is experimental. Mostly used it to get the index set but have to understand the dataset better before building it.
      - removed extract_features function form util -> model can now get all features per image, experiment manager will run retrieval/ data loading
      - added custom dataset to dataloader.py that also returns filepaths
      - perform retrieval so far extracts index features and performs pca
      
      note: pca in test run only explains ~40% of variance. is that feasible?
      
      TODO: extract features should return rf centers instead of boxes
      write matching function
      incorporate it into experiment manager
      97df86b4
  16. 15 Apr, 2020 1 commit
    • Jan Philipp Simon Langen's avatar
      Started writing retrieval code. · d6031024
      Jan Philipp Simon Langen authored
      Accumulating data from scales pyramid
      calculate receptive fields for each feature
      perform nms based on scores and receptive field boxes and keeping the topk per image
      tested running pca
      
      TODO: Find out what info we need to keep for real retrieval (image class or id?)
      Do we have multiple photos of the same landmark in our database???
      d6031024
  17. 13 Apr, 2020 1 commit
  18. 11 Apr, 2020 1 commit
  19. 08 Apr, 2020 1 commit
    • Jan Philipp Simon Langen's avatar
      -commented experiment.py code · a6fcaa49
      Jan Philipp Simon Langen authored
      -more console feedback following experiment manager and training
      -implemented logger for training
      
      TODO check if normalization should be used for keypoint training
      implement retrieval/pca
      a6fcaa49
  20. 06 Apr, 2020 1 commit
    • Jan Philipp Simon Langen's avatar
      - bug fix: gamma rescaling has to be done !first! during keypoints training · ba5b85cf
      Jan Philipp Simon Langen authored
      - bug fix: creating model/dataloader in a dict and then copying the reference during training leads to huge performance decrease. Mem requirements. Possibly model gets copied? -now building model and dataloaders within training loop
      - completly overhauled experiment manager. Now lots of checks if input is actually plausible + more flexible
      
      TODO
      comment experiment.py code
      more console feedback following experiment manager and training
      implement some kind of logger for training
      
      afterwards onwords to pca/retrieval
      ba5b85cf
  21. 04 Apr, 2020 1 commit
    • Jan Philipp Simon Langen's avatar
      -some more feedback during training · 7d9d244c
      Jan Philipp Simon Langen authored
      -proper saving/loading of model parts for keypoint/inference stages
      -built architecture for keypoint/inference
      
      -next built forward function for keypoint stage and config training params for keypoint training
      7d9d244c
  22. 03 Apr, 2020 1 commit
    • Jan Philipp Simon Langen's avatar
      -some more feedback during training · 86d70be1
      Jan Philipp Simon Langen authored
      -proper saving/loading of model parts for keypoint/inference stages
      -built architecture for keypoint/inference
      
      -next built forward function for keypoint stage and config training params for keypoint training
      86d70be1
  23. 02 Apr, 2020 1 commit
  24. 01 Apr, 2020 1 commit
  25. 30 Mar, 2020 1 commit