Deepcell-tracking

🚀 Features

<details>
<summary>Change `isbi_utils` from writing files to comparing in memory R-Ding (67)</summary>

What
* Replace writing to GT and result ISBI-style .txt files with saving an array of ISBI-style dictionaries and comparing those.
* Modifying `isbi_utils_tests` to test the new approach

Why
* Comparing in memory is quicker and allows for less chances of error
</details>


🐛 Bug Fixes

<details>
<summary>Cache the entire Python environment to speed up build times. willgraf (66)</summary>

What
* Cache the entire Python environment in the testing GitHub Action workflow.

Why
* [Drastically speed up build times](https://medium.com/ai2-blog/python-caching-in-github-actions-e9452698e98d).

</details>

<details>
<summary>Support Python 3.9 willgraf (61)</summary>


</details>


🐛 Bug Fixes

<details>
<summary>Compress `.trks` by using gzip compression when writing the tarfile. willgraf (64)</summary>

The third time's the charm! This is the same PR as 63, but removing the changes featured in 62.

This PR changes the writemode of `tarfile` to write in `w:gz`, or with gzip compression. The read mode of `tarfile` natively supports this.

Additionally, some related tests have been updated to use the `tmpdir` test fixture instead of creating their own temporary directory.
</details>

🧰 Maintenance

<details>
<summary>Bump deepcell-toolbox to 0.10.x willgraf (60)</summary>

This PR bumps the `deepcell-toolbox` dependency to 0.10.0.
</details>

🚀 Features

<details>
<summary>Update Tracking Approach and Address Outstanding Bugs MekWarrior (53)</summary>

What
* End-to-End updates for DeepCell's approach to the tracking problem in live-cell imaging. A new `Track` object is introduced to hold track information and allow for augmentation using the new dataset builder module in `deepcell-tf`. This PR introduces breaking changes related to the previous LSTM and Siamese Neural Network (SNN) model architecture. The updated `CellTracker` object is intended for use with the `inference` model and `neighborhood` encoders generated and trained by the new graph-based tracking architecture in `deepcell-tf`. Additionally, the utils related to benchmarking have been updated and improved.

Why
* These updates represent the natural evolution of the repo's approach to tracking. It addresses key needs to further enable adoption of TF2 and dramatically improves tracking speed. At the same time, it address several bugs that have been uncovered since `deepcell-tracking` was first introduced.

---

This PR should also fix several old issues:
- Fixes 13 (Removes `compute_distance`)
- Fixes 14 (Prefetches all cell features, no repeated loops)
- Fixes 35 (removes `_get_input_pairs`)

</details>

🧰 Maintenance


<details>
<summary>Fix PyPI version badge URL. willgraf (49)</summary>
</details>

<details>
<summary>Update version to 0.4.0 willgraf (58)</summary>
</details>

<details>
<summary>Create release-drafter workflow to draft releases automatically. willgraf (59)</summary>
</details>

Bugfixes

* Updated copyright year to 2021 (38)
* Manually delete `tempfile.NamedTemporaryFiles` to resolve bug on Windows (40, 41)
* Avoid integer overflow in `clean_up_annotations` (42)

Dependencies

* Use environment markers to conditionally install compatible versions of `opencv-python-headless` based on Python version (39)
* Change pinned `scikit-image` version from < 17 to >= 0.14.5 (48)

Features

- Updated `tracking.py` to consume data from dictionaries instead of lists, as required by TensorFlow 2. This is a breaking change.