Releases: Abe404/root_painter
0.3.0
To download RootPainter, click the appropriate file for your operating system from the list of assets below.
Recent changes
- Add MobileSAM inference support: segment images using pre-trained MobileSAM .pth models via Segment Folder
- Vendor MobileSAM package (no external dependencies required)
- Auto-start trainer when opening Segment Folder in workstation mode
- File picker now accepts both .pkl (UNet) and .pth (MobileSAM) model files
Ubuntu Workstation
Which file do I need? Run nvidia-smi in a terminal to see your GPU name, then pick the matching download.
| GPU | Download |
|---|---|
| GTX 1660, RTX 2060–2080, RTX 3060–3090, RTX 4060–4090 | RootPainterWorkstation_0.3.0_Ubuntu_CUDA128_GTX1660_to_RTX4090.AppImage |
| RTX 5090, 5080, 5070 Ti (Blackwell) | RootPainterWorkstation_0.3.0_Ubuntu_CUDA128_RTX50.AppImage |
Server install
pip install root-painter-trainer==0.3.0
0.2.29
To download RootPainter, click the appropriate file for your operating system from the list of assets below.
Recent changes
- Fix uint16 TIFF images displaying as blank: #191
- Add
--maxbatchsizeargument for limiting GPU memory on shared systems: #193 - Extend dataset no longer includes hidden files: #173
- Warn when starting training while a model is already training: #136
- Trainer status check via instruction for workstation mode: ca341bf
- Warn when opening project without server running: 27f699f
- Default sync dir to ~/root_painter_sync with yes/no dialog: 40e2294
- Fix Windows workstation build: c50499c
Ubuntu Workstation
Which file do I need? Run nvidia-smi in a terminal to see your GPU name, then pick the matching download.
| GPU | Download |
|---|---|
| GTX 1660, RTX 2060–2080, RTX 3060–3090, RTX 4060–4090 | RootPainterWorkstation_0.2.29_Ubuntu_CUDA128_GTX1660_to_RTX4090.AppImage |
| RTX 5090, 5080, 5070 Ti (Blackwell) | RootPainterWorkstation_0.2.29_Ubuntu_CUDA128_RTX50.AppImage |
Server install
pip install root-painter-trainer==0.2.29.0
Custom PyTorch wheels (CUDA 12.8, sm_75/80/86/89)
Minimal PyTorch 2.7.1 + torchvision 0.22.1 wheels for RootPainter workstation.
Supports: GTX 1660 (Turing), RTX 2060-2080, RTX 3060-3090, RTX 4060-4090.
CUDA architectures: sm_75, sm_80, sm_86, sm_89.
Custom PyTorch wheels (CUDA 12.8, sm 12.0)
Minimal PyTorch + torchvision wheels for RootPainter workstation builds.
PyTorch ref: v2.7.1
torchvision ref: v0.22.1
CUDA arch: 12.0 (Blackwell)
Built without: nccl, cusparse, cusparselt, cufft, cusolver, distributed, triton, flash-attention
These wheels are used by build_workstation_ubuntu_cuda128.sh as a fallback when no local wheels exist in ./dist/.
To rebuild locally: ./build_custom_torch.sh
0.2.28
To download RootPainter, click the appropriate file for your operating system from the list of assets below.
New in this release: RootPainter Workstation — a version with the trainer built-in, so you can annotate and train on the same machine without installing the server separately.
Recent changes
- RootPainter Workstation edition — integrated trainer and painter in a single application (Windows, macOS, Ubuntu)
- Mac training support (Apple Silicon M1/M2 via MPS): #86
- CPU inference support: #118
- Fix loading CUDA-trained models on Mac MPS: eceb3a4
- Fix masking not working: #182
- Increase maximum image size: #134
- Fix np.int deprecation: #133
- Keyboard shortcuts moved to Help menu: #81
- Handle JPEG quality bug fix: 86572af
Ubuntu Workstation
Which file do I need? Run nvidia-smi in a terminal to see your GPU name, then pick the matching download.
| GPU | Download |
|---|---|
| GTX 1660, RTX 2060–2080, RTX 3060–3090, RTX 4060–4090 | RootPainterWorkstation_0.2.28_Ubuntu_CUDA128_GTX1660_to_RTX4090.AppImage |
| RTX 5090, 5080, 5070 Ti (Blackwell) | RootPainterWorkstation_0.2.28_Ubuntu_CUDA128_RTX50.AppImage |
Download the AppImage, right click, change properties to make it executable, then double-click it or run:
chmod +x RootPainterWorkstation_*.AppImage
./RootPainterWorkstation_*.AppImage
Server install
pip install root-painter-trainer==0.2.28.1
0.2.27
To download the RootPainter client. Click the appropriate file for your operating system from the list of assets below.
Recent changes in this release include:
- Masks resized to work for larger images #103
- Optionally enhance image contrast: #102
- Plot metrics in terms of interaction time or clicks: #102
- View current brush size in painter interface: #100
- View model name/number in corrective-metrics plot (hover over point) and in output CSV #98
- Handle P-mode images when creating dataset: #94
- Fixing bug with 'view image in context': #93 and #92
- Add option to open sync directory (extras menu): #91
- Making it more convenient to re-specify dataset location: #90
- More sensible defaults for directory location when creating projects: #63
- Resize image option: fa815e1
- Improved exception logging for client: a3175e8
Server install:
This can be done by cloning the server source code again or installing with pip:
pip install root-painter-trainer==0.2.27.0
0.2.26.pre
This is a work in progress release for development and testing purposes. The latest stable release is available here: https://github.com/Abe404/root_painter/releases/tag/0.2.25
0.2.25
To download the RootPainter client. Click the appropriate file for your operating system from the list of assets below.
Recent changes in this release include:
- New functionality to assign corrections to segmentations available from the extras menu, with contributions from @rohanorton in #83 . This improves semi-automatic segmentation workflows and enables easier comparison of the properties of the corrected segmentations to the predicted segmentations which is useful for model validation.
- Now possible to directly output segmentations in a format compatible with RhizoVision Explorer
- More robust handling of dataset renaming and moving (you can now move and rename datasets already used in projects without RootPainter crashing when it is reopened).
Server install:
The RhizoVision segmentation option included in this client also requires that you also update the server to the latest version. This can be done by cloning the server source code again or installing with pip:
pip install root-painter-trainer==0.2.25.1
Contributors
Assets 6
0.2.24
To download the RootPainter client. Click the appropriate file for your operating system from the list of assets below.
Recent changes in this release include:
- A fix for a bug with viewing image in context caused by image not being divided evenly by patch size.
- Conversion to ubyte before conversion to pixmap in client to avoid some 64bit PNG images not showing.
Full Changelog: 0.2.23...0.2.24
0.2.23
Note this is no longer the latest release. Latest release available at: https://github.com/Abe404/root_painter/releases
In this release, predicted area, corrected area, and the difference between them (area error) are now available from the metrics plot, providing feedback on model performance in terms of under or over-estimation of area during interactive model training.
Full Changelog: 0.2.22...0.2.23
