Skip to content

Conversation

@tlarcher
Copy link
Collaborator

@tlarcher tlarcher commented Apr 9, 2025

Update v2.2.0

📝 Changelog

  • GLC25

    • Added working example (train, transfer, prediction). Inference mode will be provided after the end of the Kaggle challenge.
    • Added unit tests
    • Pre-computed approx mean and std for each modality
    • Added custom transform function and pre-computed moments (mean & std) on the Satellite patches (cf. dataset/geolifeclef-2025/Stats/)
  • Toolbox

    • compute_mean_std_iteratively_from_sample.py
      • Added .tiff files support
      • BugFix: Now still computes mean and std if any dataset sample is full of NaN values, instead of outputting 0

✅ Checklist

  • Lint and tests pass locally with my changes
  • I've added necessary documentation

tlarcher and others added 19 commits November 22, 2024 15:20
…tes the mean and std even if some data elements are full of NaN
- compute_mean_std_iteratively_from_sample.py: modified tiff loading function to permute axis to match img loading function
- fixed typos
…transform values accordingly.

- .gitignore: added data folders with typo in name
- glc25_cnn_multimodal_ensemble.py: added necessary class atributes to allow prediction files export
- Modified GLC25 to correctly provide multilabel and multiclass behaviors. Previously, multilabel would only give 1 one in the one-hot vector (so does GLC24 at the moment git status)
Added GLC25 example and all the related tests.
Improved compute_mean_std_iteratively_from_sample.py
updated libgeos-dev system package version
Updated libgeos-dev system package version
Updated libgeos-dev system package version
Deleted version specification for libgeos-dev
…ersions couldn't be found by GitHub package manager
@tlarcher tlarcher self-assigned this Apr 9, 2025
@tlarcher tlarcher added bug Something isn't working documentation Improvements or additions to documentation new feature labels Apr 9, 2025
tlarcher and others added 2 commits April 9, 2025 15:09
* Temporary removed call for quantile_normalize

* WiP: Added transform methods ready to be called in the pipeline. Need to verify well behavior of pre_compute of linear quantiles method and lint file

* Added custom transforms for GLC25. Changed GLC25 main file weights dir directory to avoid re-downloading the MME pre-trained weight over again. Renamed GLC25 Stats folder

* Linting on GLC25 files
@tlarcher tlarcher merged commit 8078987 into main Apr 16, 2025
1 check passed
tlarcher added a commit that referenced this pull request Jul 8, 2025
* Updated setup.py for v2.1.2

* Added GLC25 related stuff

* Added GLC25 stats and data Readme

* Corrected typos

* Updated compute_mean_std_iteratively_from_sample.py so it still computes the mean and std even if some data elements are full of NaN

* Added GLC25 test stats

* Added GLC25 satellite test stats.
- compute_mean_std_iteratively_from_sample.py: modified tiff loading function to permute axis to match img loading function
- fixed typos

* Computed channel-wise satellite patches moments and updated the data transform values accordingly.
- .gitignore: added data folders with typo in name

* Confirmed GLC25 is running for training, transfert, inference.
- glc25_cnn_multimodal_ensemble.py: added necessary class atributes to allow prediction files export

* GLC25 config: restored default values.

* Added GLC25 unit tests.
- Modified GLC25 to correctly provide multilabel and multiclass behaviors. Previously, multilabel would only give 1 one in the one-hot vector (so does GLC24 at the moment git status)

* Linting

* Added GLC25 in test_examples and ran it: OK

* Update test.yml

updated libgeos-dev system package version

* Update dispatch-test-examples.yml

Updated libgeos-dev system package version

* Update dispatch-test.yml

Updated libgeos-dev system package version

* Update dispatch-test.yml

Deleted version specification for libgeos-dev

* Removed libgeos-dev version constraint in GitHub workers as several versions couldn't be found by GitHub package manager

* Updated setup.py. Fixed typo

* Glc25 (#78)

* Temporary removed call for quantile_normalize

* WiP: Added transform methods ready to be called in the pipeline. Need to verify well behavior of pre_compute of linear quantiles method and lint file

* Added custom transforms for GLC25. Changed GLC25 main file weights dir directory to avoid re-downloading the MME pre-trained weight over again. Renamed GLC25 Stats folder

* Linting on GLC25 files
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

bug Something isn't working documentation Improvements or additions to documentation new feature

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants