v3.2.0

What's Changed

• v3.1.0 by @tkswanson in #194
• check for checkpoint path first by @tkswanson in #196
• 197 sort visualize by @tkswanson in #199
• 162 export coco by @tkswanson in #202
• add -d detect_only option to command line by @tkswanson in #203
• fix cache path bug by @tkswanson in #204
• update fro cometml 3.54, add optional dependency in toml by @tkswanson in #206
• typecast datetime to ensure correct transmission to animl-r by @tkswanson in #208
• check device first before setting pin_memory to true by @tkswanson in #210
• switch function to fillna by @tkswanson in #211
• fix bug with Path version of model_path by @tkswanson in #213
• fix export_timelapse for 3.1 by @tkswanson in #214
• add get_device_onnx by @tkswanson in #217
• fix inputs for onnx and series by @tkswanson in #219
• 182 train test split by @tkswanson in #220
• Dev by @tkswanson in #215
• 222 miew onnx by @tkswanson in #223
• add user_set for get_onnx_device, change get_device to get_torch_device by @tkswanson in #225
• convert none empty to empty df by @tkswanson in #227
• 232 onnx providers by @tkswanson in #233
• fix single_clasification to handle predictions_raw indexing correctly by @tkswanson in #234
• animl 3.2.0 by @tkswanson in #224

Full Changelog: v3.1.0...v3.2.0

3.1.0 - Video and Onnx

update to python >=3.12
add .onnx support for MD and classifiers (onnxruntime >=1.23)
load_classifier also loads class_list, either from file or from model metadata
frames are now extracted on the fly, ImageGenerator -> ManifestGenerator
WorkingDir updated to remove Frames subfolder, Data subfolder
single_classification updated to handle video
update plot_box to optionally return image or save
add onnx tests

v3.0.0

Animl 3.0.0 is a complete overhaul of the API including:

• reorganization of detection and classification modules
• reordering and addition of arguments
• overhaul of model training functions and hyperparameter options
• addition of reid module
• fix visualization functions
• rename of link/sort to export
• generator for megadetector
• added support for MDv6 and MDv1000
• add tools to download models

v2.0.0

Modifications:

• improve sequence classification
• add custom yolo detection
• skip corrupted videos in extract_frames()
• add file_col to split() to prevent data leakage
• simplify matchypatchy api
• add optimizer and schedular kwargs to save_model
• add device as model class attribute
• add parameter and hyperparameter configuration options for training
• add cache_dir kwarg for TrainGenerator and train_dataloarder
• add typing
• flake8

Deletions:

• remove exiftools as a dependency
• remove detect_mp
• remove utils/dataloaders.py
• remove all but letterbox() and exif_transpose() from utils/augmentations

New Contributors

• @mamani828 made their first contribution in #85

Full Changelog: v1.4.2...v2.0.0

Animl 1.4.2

v1.4.2

Merge branch 'main' of github.com:conservationtechlab/animl-py

Animl-py v1.3.0

Release Jul 25, 2024
Major Changes:

• PyTorch backend (remove TF)
• Train Classifiers
• Efficiency and speed improvements
• Compatibility with R package

animl-py v1.1.0

Update README.md