Fix version 0.2.0-dev3 and later

In this MR, version 0.2.0-dev3 (and later) is fixed:

• We use DPIAdjusting transform as it is a key component when training on images of very different sizes
• We compute the real mean and std values on the train set (we do not use default values from ImageNet = revert a5f7b5ab)
• The padding is on the right and bottom of the image (= revert part of f57bc667 for now)
• Two parameters files are created during training: one for training details (used to reproduce/continue the experiment), one for inference (that can be directly used by Arkindex DAN worker)

assigned to @mboillet

added 1 commit

• e6633338 - Fix tests models

Compare with previous version

added 1 commit

• bc055c72 - Fix tests

Compare with previous version

added 6 commits

• 75c828f4 - Apply bb4b9f4
• e473fbe3 - Correctly typing parameters
• 905a71db - Do not download images if they already exist
• 5df8dd8b - Apply be01536c
• 57aa0063 - Enable to use GPU for teklia_dan_predict
• 0d4ac844 - Store absolute path of images

Compare with previous version

added 2 commits

• dfd6835b - Fix input tensor used during prediction
• 5d35d16d - Set default value to None for --image_max_width during prediction

Compare with previous version

added 1 commit

• 8ba39efd - Apply 2bb85c50

Compare with previous version

added 6 commits

• dad76649 - Apply 8112af86
• 9d99164b - Apply d580a769
• 35d3b1fc - Apply d2760687
• 82fe02a1 - Apply 5f7a5c2f
• 59a81d9b - Use a specific version of griffe to fix the documentation build
• a33ee4bb - Apply 2330f7ee

Compare with previous version

added 9 commits

• c0ac48fa - Version 0.2.0-dev3
• 2e67db27 - Apply 6679f473
• 79c51d78 - Apply f0a6e38c
• 20dbc910 - Apply 65177a4b
• dc482e24 - Apply cb58891e
• 95b168cf - Apply a633050e
• cddebe2f - Apply 4c1ccf10 + generate a inference_parameters.yml file
• 9d2d6bf5 - Fix worker version filters
• a6d52a94 - Remove typo in documentation

Compare with previous version

added 2 commits

• 14a829a4 - Apply 774887b1
• dc255d49 - Apply deab8965

Compare with previous version

added 5 commits

• 3ac82063 - Document batch prediction arg
• 45c7e802 - Apply 803610d3
• ba61688a - Fix warning admonition
• ee994ff3 - 30613c99
• bac1de3c - Apply 1956acc9

Compare with previous version

I have applied all commits until 1956acc9 (current main version). I am now testing this version during a real training on Jean Zay

added 62 commits

• bac1de3c...1956acc9 - 52 commits from branch main
• 70e99d77 - Fix padding to bottom-right
• f60631d2 - Apply 78e6f601 + keep DPIAdjusting
• 3dfe694e - Fix data augmentations
• 55698f0c - Load images using Torch
• 327fbd97 - Fix tests models
• 1392b8f4 - Fix tests
• 59ab6de9 - Apply 2bb85c50
• 4fe27870 - Apply f0a6e38c
• bff8b5e7 - Apply deab8965
• 805ac307 - Fix tests model

Compare with previous version

added 1 commit

• f627d8e6 - Remove popp training code

Compare with previous version

added 1 commit

• 8dace579 - Fix rebase

Compare with previous version

I have rebased on main. I will now test this version during a real training on Jean Zay.

changed the description

changed milestone to %ML Prod - Summer 2023

added P1 label

changed title from Draft: Fix dev3 to Draft: Fix version 0.2.0-dev3 and later

requested review from @yschneider

This code works as expected (see yellow curve below), it can be reviewed and merged

marked this merge request as ready

added 1 commit

• 40e16db2 - Apply review suggestions

Compare with previous version

resolved all threads

approved this merge request

merged