Recent Releases of Zamba
Zamba - v2.5.0
Note: This version removes support for python 3.8 and 3.9.
What's Changed
- Fix ffmpeg install on macos by @ejm714 in https://github.com/drivendataorg/zamba/pull/329
- Fix ModelManager import in docs by @klwetstone in https://github.com/drivendataorg/zamba/pull/325
- Add code of conduct by @pjbull in https://github.com/drivendataorg/zamba/pull/330
- Pin numpy < 2 by @pjbull in https://github.com/drivendataorg/zamba/pull/335
- Replace maximize-build-space action by @jayqi in https://github.com/drivendataorg/zamba/pull/336
- Upgrade the required python version to python>=3.11 by @westford14 in https://github.com/drivendataorg/zamba/pull/337
New Contributors
- @westford14 made their first contribution in https://github.com/drivendataorg/zamba/pull/337
Full Changelog: https://github.com/drivendataorg/zamba/compare/v2.4.1...v2.5.0
Biosphere - Terrestrial Wildlife
- Python
Published by ejm714 over 1 year ago
Zamba - v2.3.0
Model release
- Adds a depth estimation module for predicting the distance between animals and the camera (PR #247). This model comes from one of the winning solutions in the Deep Chimpact: Depth Estimation for Wildlife Conservation machine learning challenge hosted by DrivenData.
Biosphere - Terrestrial Wildlife
- Python
Published by ejm714 over 3 years ago
Zamba - v2.2.1
- Turn off showing local variables in Typer's exception and error handling (PR #237)
- Fixes bug where the column order was incorrect for training models when the provided labels are a subset of the model's default labels (PR #236)
Biosphere - Terrestrial Wildlife
- Python
Published by ejm714 over 3 years ago
Zamba - v2.2.0
Model releases and new features
- The default
time_distributedmodel (African species classification) has been retrained on over 250,000 videos. This 16x increase in training data significantly improves accuracy. This new version replaces the previous one. (PR #226, PR #232) - A new default model option is added:
blank_nonblank. This model only does blank detection. This binary model can be trained and finetuned in the same way as the species classification models. This model was trained on both African and European data, totaling over 263,000 training videos. (PR #228) - Detect if a user is training in a binary model and preprocess the labels accordingly (PR #215)
Bug fixes and improvements
- Add a validator to ensure that using a model’s default labels is only possible when the species in the provided labels file are a subset of those (PR #229)
- Refactor the logic in
instantiate_modelfor clarity (PR #229) - Use pqdm to check for missing files in parallel (PR #224)
- Set
model_namebased on the provided checkpoint so that user-trained models use the appropriate video loader config (PR #221) - Leave
data_diras a relative path (PR #219) - Ensure hparams yaml files get included in the source distribution (PR #210)
- Hold back setuptools so mkdocstrings works (PR #207)
- Factor out
get_cached_array_path(PR #202)
Biosphere - Terrestrial Wildlife
- Python
Published by ejm714 over 3 years ago
Zamba - v2.1.0
- Retrains the time distributed species classification model using the updated MegadetectorLite frame selection (PR #199)
- Replaces the MegadetectorLite frame selection model with an improved model trained on significantly more data (PR #195)
Biosphere - Terrestrial Wildlife
- Python
Published by ejm714 almost 4 years ago
Zamba - v2.0.4
- Pins
thopto an earlier version (PR #191) - Fixes caching so a previously downloaded checkpoint file actually gets used (PR #190, PR #194)
- Removes a lightning deprecation warning for DDP (PR #187)
- Ignores extra columns in the user-provided labels or filepaths csv (PR #186)
Biosphere - Terrestrial Wildlife
- Python
Published by ejm714 almost 4 years ago
Zamba - v2.0.1
Releasing to pick up #167 and #169.
- PR #169 fixes error in splitting data into train/test/val when only a few videos.
- PR #167 refactors yolox into an object_detection module
Other documentation fixes also included.
Biosphere - Terrestrial Wildlife
- Python
Published by pjbull over 4 years ago
Zamba - v2.0.0
Releases new version v2 of the zamba package including:
- New and improved models
- Additional species and habitats
- Model finetuning and training
Previous model: Machine learning competition
The algorithms used by zamba v1 were based on the winning solution from the Pri-matrix Factorization machine learning competition, hosted by DrivenData. Data for the competition was provided by the Chimp&See project and manually labeled by volunteers. The competition had over 300 participants and over 450 submissions throughout the three month challenge. The v1 algorithm was adapted from the winning competition submission, with some aspects changed during development to improve performance.
The core algorithm in zamba v1 was a stacked ensemble which consisted of a first layer of models that were then combined into a final prediction in a second layer. The first level of the stack consisted of 5 keras deep learning models, whose individual predictions were combined in the second level of the stack to form the final prediction.
In v2, the stacked ensemble algorithm from v1 is replaced with three more powerful single-model options: time_distributed, slowfast, and european. The new models utilize state-of-the-art image and video classification architectures, and are able to outperform the much more computationally intensive stacked ensemble model.
New geographies and species
zamba v2 incorporates data from western Europe (Germany). The new data is packaged in the pretrained european model, which can predict 11 common European species not present in zamba v1.
zamba v2 also incorporates new training data from 15 countries in central and west Africa, and adds 12 additional species to the pretrained African models.
Retraining flexibility
Model training is made available zamba v2, so users can finetune a pretrained model using their own data to improve performance for a specific ecology or set of sites. zamba v2 also allows users to retrain a model on completely new species labels.
Biosphere - Terrestrial Wildlife
- Python
Published by pjbull over 4 years ago