roadmap.md

A development roadmap for DeepLabCut

📢 ⏳ 🚧

deeplabcut 2.2: multi-animal pose estimation and tracking with DeepLabCut

• alpha testing complete (early May 2020)
• beta release: 2.2.b5 on 5 / 22 / 20 😄
• beta release: 2.2b8 released 9/2020 😄
• beta release 2.2b9 (rolled into 2.1.9 --> candidate release, slotted for Oct 2020)
• 2.2rc1
• 2.2rc2
• 2.2rc3
• Manuscript Lauer et al 2021 https://www.biorxiv.org/content/10.1101/2021.04.30.442096v1
• full 2.2 stable release

real-time module with DEMO for how to set up on your camera system, integration with our [Camera Control Software](https://github.com/AdaptiveMotorControlLab/Camera_Control)

• Integration with Bonsai completed! See: https://github.com/bonsai-rx/deeplabcut
• Integreation with Auto-pi-lot. See: https://auto-pi-lot.com/
• DeepLabCut-live! released Aug 5th, 2020: preprint & code: https://www.biorxiv.org/content/10.1101/2020.08.04.236422v1
• DeepLabCut-live! published in eLife

DeepLabCut Model Zoo: a collection of pretrained models for plug-in-play DLC and community crowd-sourcing.

• BETA release with 2.1.8b0: http://www.mousemotorlab.org/dlc-modelzoo
• full release with 2.1.8.1 http://www.mousemotorlab.org/dlc-modelzoo
• Manuscript forthcoming!
• new models added; horse, cheetah
• TopView_Mouse model
• contribution module

DeepLabCut GUI and DeepLabCut-core:

• to make DLC more modular, we will move core functions to https://github.com/DeepLabCut/DeepLabCut-core
• DLC-core depreciated, and core is now simply pip install deeplabcut GUI is with pip install deeplabcut[gui]
• new GUI for DeepLabCut; due to extended issues with wxPython, we will be moving to release a napari plugin https://github.com/napari/napari
• New project management GUI
• tensorflow 2.2 support in DeepLabCut-core: DeepLabCut#601
• DeepLabCut-Core to be depreciated; TF2 will go into main repo.
• TF2 support while also maintaining TF1 support until 2022.
• Web-based GUI for labeling --> Colab training pipeline for users (full no-install DLC)

General NN Improvements:

• EfficientNet backbones added (currently SOTA on ImageNet). https://openaccess.thecvf.com/content/WACV2021/html/Mathis_Pretraining_Boosts_Out-of-Domain_Robustness_for_Pose_Estimation_WACV_2021_paper.html https://github.com/DeepLabCut/DeepLabCut/commit/96da2cacf837a9b84ecdeafb50dfb4a93b402f33
• New multi-fusion multi-scale networks; DLCRNet_ms5

General Enhancements:

• DeepLabCut PyTorch model zoo
• DLC-CookBook v0.1
• DLC BLog for releases and user-highlights
• New Docker containers into main repo / linked to Docker hub and repo(s)
• 3D >2 camera support