Closed
Milestone
ML Prod - Next
Unstarted Issues (open and unassigned)
2
Ongoing Issues (open and assigned)
0
Completed Issues (closed)
22
- Extract data without any transcription
- Charset should only include training characters
- Error in _valid_image_format
- Filter entities by name when extracting data from Arkindex
- Add an argument for entity tokens during prediction
- Do not permute input tensor during prediction
- Load model in eval mode by default during prediction
- Remove unused decoder attributes
- Enable to use GPU for teklia_dan_predict
- Resize images during extraction
- Add predicted text and confidence associated
- Add predicted objects to predict command
- Prediction on a test dataset
- Compute confidence scores by char, word or line
- Support mlflow logging for line training and properly document it
- Fix the resize of images to download when they are too big
- Offset issue during dataset extraction using start_token only
- Document model inference
- Document Jean-Zay training
- Document a generic workflow using this repo
- Refactor generate subcommand
- Documentation improvements
Loading
Loading
Loading