Skip to content
Snippets Groups Projects
CHANGELOG.md 2.81 KiB
Newer Older
  • Learn to ignore specific revisions
  • # MLCVZoo mlcvzoo_yolox module Versions:
    
    6.2.1 (2024-08-05):
    ------------------
    Hotfix: Remove debug timing of the model
    
    6.2.0 (2024-06-07):
    ------------------
    Fix checkpoint name of trained yolox models
    - Use leading zeros for correct order of file-paths
    - Additional Bugfix: Correctly state onnx extra 
    
    6.1.0 (2024-06-07):
    ------------------
    Implement and use API changes introduced by mlcvzoo-base version 6.0.0
    
    6.0.1 (2023-05-03):
    ------------------
    Python 3.10 compatibility
    
    6.0.0 (2023-02-14):
    ------------------
    Implement API changes introduces by mlcvzoo-base version 5.0.0
    - Remove detector-config and use the feature of the single ModelConfiguration
    - Remove duplicate attributes
    
    5.3.0 (2023-02-08):
    ------------------
    - Include YOLOX multi-gpu training capabilities
    - Update poetry lock file
    -
    5.2.0 (2023-01-25):
    ------------------
    - Upgrade TensorRT version from 8.2.3.0 to 8.4.2.4
      - Increment version in mlcvzoo_yolox/pyproject.toml
    - Add tests for the conversion of different yolox versions (s, m, l, x) to TensorRT models
      - The higher TensorRT version yields no reduction in memory costs,
        a lack of VRAM on the GPU still results in an error
      - Update torch2trt to latest stable commit
    - Fix post-processing bug for yolox predictions
      - Add the third-party module boxes.py where the bug resides
      - Always choose the first detection (highest NMS score) if there are several available
    
    5.1.1 (2022-11-10):
    ------------------
    Remove dependency on backports.strenum
    
    5.1.0 (2022-09-09):
    ------------------
    - Ensure ConfigBuilder version 7 compatibility
    - Remove call to yolox Trainer.after_epoch(...)
      - It generated duplicated checkpoints
      - This removes the possibility to run an evaluation during training
      - For evaluating model checkpoints after training the mlcvzoo-modeltrainer
        (mlcvzoo-util module) is recommended
    
    5.0.0 (2022-08-08):
    ------------------
    - Adapt to mlcvzoo-base 4.0.0
    
    4.0.2 (2022-07-11):
    ------------------
    Prepare package for PyPi
    
    4.0.1 (2022-06-29):
    ------------------
    Fix num_classes bug:
    - Ensure that the model in yolox gets initialized with the correct number of classes
    
    4.0.0 (2022-06-13):
    ------------------
    Refactor and update:
    - Update to yolox version 0.3.0
    - Refactor initialization of yolox experiments in the MLCVZoo
      - Define the default parameters for the yolox versions (nano, s, m, l ...) in a lookup dictionary
      - Don't use an extra .py experiment file, but add a config entry that allows to
        define parameters that should be overwritten for the experiments
      - Remove unneeded configuration attributes
    - Fix num_classes method
      - Add missing decorator
      - Don't use deprecated attribute
    
    3.0.0 (2022-05-16):
    ------------------
    Use new features from AnnotationClassMapper that have been added with mlcvzoo_base v3.0.0
    
    2.0.0 (2022-04-05)
    ------------------
    - initial release of the package