Releases: t29mato/AutoLineDigitizer
Releases · t29mato/AutoLineDigitizer
Release v0.4.0
AutoLineDigitizer v0.4.0
New Features
- Line Model selection: Switch between General and Battery (fine-tuned) models via dropdown in sidebar
- Axis Model dropdown: Axis detection model (ChartDete) shown as dropdown in sidebar for future extensibility
- Manual model import: Import Models button with file picker for proxy/firewall environments where auto-download fails
- SHA256 hash verification: Validates imported model files to prevent loading corrupted/wrong files
Improvements
- EasyOCR models bundled: OCR models included in the app — no network access needed for axis label reading
- Proxy/SSL support: Robust fallback for corporate proxy environments (manual browser download + import)
- UI reorganization: Model loading status moved to sidebar under each model dropdown; extraction status shown in main panel
- Text wrapping: Status messages and download help UI wrap properly within sidebar width
- Button simplification: "Open Image (or Cmd+V)" merged into a single button
Bug Fixes
- Fix
init_cfgpretrained weight download in both LineFormer and ChartDete configs (caused SSL errors behind proxies) - Fix model loading crash when using parallel threads (
mmcv.Config.fromfilethread-safety issue) - Fix download help UI accumulation on repeated model switches
- Fix EasyOCR model download failure in bundled app
Download
- macOS (Apple Silicon):
AutoLineDigitizer-macOS.zip - Windows:
AutoLineDigitizer-Windows.zip
Installation
macOS
- Download and unzip
AutoLineDigitizer-macOS.zip - Move
AutoLineDigitizer.appto Applications - Right-click > "Open" (required for first launch)
- Models will be downloaded automatically on first launch
Windows
- Download and unzip
AutoLineDigitizer-Windows.zip - Run
AutoLineDigitizer\AutoLineDigitizer.exe - Models will be downloaded automatically on first launch
See Manual Model Download for proxy environments.
Release v0.3.1
AutoLineDigitizer v0.3.1
What's Changed (since v0.3.0)
Bug Fix
- Fixed SSL certificate error: Model downloads from HuggingFace now work correctly in the bundled app (previously failed with SSL EOF error)
New
- Version display: App version is now shown at the bottom of the Settings panel
Full Changelog (since v0.2.0)
Line Model Selection
- Multiple line extraction models: Choose between General and Battery-specialized models from the Settings panel
- Battery (iter_5000): Best for accurate line counting with fewer false positives (recommended for charge/discharge curves)
- Battery (best_segm): Best for precise mask segmentation boundaries
- Auto-download from HuggingFace: Fine-tuned models are downloaded automatically on first use with progress bar
Improved Point Extraction
- Replaced the old extraction algorithm with a median-based centerline algorithm
- More accurate point placement along detected lines
UX Improvements
- Immediate results: Extracted points are displayed right after line extraction
- Faster model switching: Axis detection results are preserved when switching line models
- Paste hint: "or Cmd+V to paste" text shown next to Open Image button
Download
- macOS (Apple Silicon):
AutoLineDigitizer-macOS.zip - Windows:
AutoLineDigitizer-Windows.zip
Installation
macOS
- Download and unzip
AutoLineDigitizer-macOS.zip - Move
AutoLineDigitizer.appto Applications - On first launch, macOS will show a warning: "AutoLineDigitizer.app cannot be opened because the developer cannot be verified."
- Click Cancel (not "Move to Trash")
- Go to System Settings → Privacy & Security → scroll down and click "Open Anyway"
- Or: Right-click the app → Open → click Open in the dialog
- Models will be downloaded automatically on first launch
Windows
- Download and unzip
AutoLineDigitizer-Windows.zip - Run
AutoLineDigitizer\AutoLineDigitizer.exe - Models will be downloaded automatically on first launch
Release v0.3.0
AutoLineDigitizer v0.3.0
What's New
Line Model Selection
- Multiple line extraction models: Choose between General and Battery-specialized models from the Settings panel
- Battery (iter_5000): Best for accurate line counting with fewer false positives (recommended for charge/discharge curves)
- Battery (best_segm): Best for precise mask segmentation boundaries
- Auto-download from HuggingFace: Fine-tuned models are downloaded automatically on first use with progress bar
Improved Point Extraction
- Replaced the old extraction algorithm (fixed-interval sampling + component analysis + linear interpolation) with a median-based centerline algorithm
- More accurate point placement along detected lines
- No more missing points at thick or branching regions of the mask
UX Improvements
- Immediate results: Extracted points are displayed right after line extraction, without waiting for axis detection
- Faster model switching: Axis detection results are preserved when switching line models (no re-detection needed)
- Paste hint: "or Cmd+V to paste" text shown next to Open Image button
README
- Replaced Streamlit setup instructions with desktop app download guide
- Added macOS Gatekeeper warning workaround instructions
- Added Intel Mac compatibility note (Apple Silicon only for now)
Download
- macOS (Apple Silicon):
AutoLineDigitizer-macOS.zip - Windows:
AutoLineDigitizer-Windows.zip
Installation
macOS
- Download and unzip
AutoLineDigitizer-macOS.zip - Move
AutoLineDigitizer.appto Applications - On first launch, macOS will show a warning: "AutoLineDigitizer.app cannot be opened because the developer cannot be verified."
- Click Cancel (not "Move to Trash")
- Go to System Settings → Privacy & Security → scroll down and click "Open Anyway"
- Or: Right-click the app → Open → click Open in the dialog
- Models will be downloaded automatically on first launch
Windows
- Download and unzip
AutoLineDigitizer-Windows.zip - Run
AutoLineDigitizer\AutoLineDigitizer.exe - Models will be downloaded automatically on first launch
Release v0.2.0
AutoLineDigitizer v0.2.0
What's New
New Features
- Clipboard paste support: Load images directly with Cmd+V (macOS) / Ctrl+V (Windows)
- Auto-extraction on image load: Automatically runs line extraction when a new image is loaded (including 2nd+ images)
- Instant downsampling: Inference results are cached, allowing real-time parameter adjustments without re-running the model
- Auto-reload on settings change: Changing downsampling method, max points, or sort order instantly updates results
- Responsive image display: Images now fill the available width for better visibility
- Digitizer tool export: Export extracted data directly to StarryDigitizer (.zip) and WebPlotDigitizer (.tar) for fine-tuning
- External links: Quick-access buttons to open StarryDigitizer and WebPlotDigitizer
UI Improvements
- Simplified sort dropdown labels (e.g. "Detection Order" instead of "Original (Detection Order)")
- Removed unnecessary "Show visualization" checkbox
- Moved status indicators (Models loaded, Axis detection) to main panel
- Default sort changed to "Mean Y (High → Low)"
- Max Points slider now displays current value (default: 20, max: 100)
- Top-aligned image layout instead of vertical centering
- Redesigned export section with "Adjust in Digitizer" heading, grouped by tool
Build & CI
- Cross-platform build support (macOS Apple Silicon + Windows)
- GitHub Actions release workflow with bundled EasyOCR models
- Auto-download models on first launch
- Pinned flet 0.24.1 and mmcv-full 1.7.2 for stable builds
- Fixed mmcv build issues (setuptools compatibility,
--no-build-isolation) - Fixed ChartDete mmcv 2.x version check compatibility
Documentation
- Added demo video to README
- Updated model weights section with links to source repositories
Download
- macOS (Apple Silicon):
AutoLineDigitizer-macOS.zip - Windows:
AutoLineDigitizer-Windows.zip
Installation
macOS
- Download and unzip
AutoLineDigitizer-macOS.zip - Move
AutoLineDigitizer.appto Applications - Right-click > "Open" (required for first launch)
- Models will be downloaded automatically on first launch
Windows
- Download and unzip
AutoLineDigitizer-Windows.zip - Run
AutoLineDigitizer\AutoLineDigitizer.exe - Models will be downloaded automatically on first launch
v0.1.0 - Initial Release
AutoLineDigitizer
Automatic line extraction from chart images.
Features
- LineFormer for line extraction
- ChartDete + EasyOCR for automatic axis detection
- Export to StarryDigitizer and WebPlotDigitizer formats
Demo
See the attached video for a demonstration of the Streamlit app.
Requirements
- Python 3.8+
- PyTorch
- mmcv-full, mmdet
- See requirements.txt for full list
Model Weights (CI)
Model weights for AutoLineDigitizer builds. Do not delete - required by CI and app auto-download.