GitLab

- Add PPHGNetV2_B4 backbone to the list of supported backbones
- Introduce new OCR model configuration for PP-OCRv5 with PPHGNetV2_B4
- Update existing model configurations to use the new backbone
- Modify RNN neck to support input with H > 1
- Adjust batch size for inference

- Add new PP-OCRv5 detection and recognition models
- Update arch_config.yaml with new model architectures
- Modify models_config.yml to include PP-OCRv5 models for ch_lite configuration- Change dictionary file for ch_lite to ppocrv5_dict.txt

- Add logger info for each batch processed
- Include batch number and page count in log message

- Adjust the threshold for considering tables inside other tables from2 to 3
- Add support for custom formula delimiters through user configuration
- Pin pdfminer.six to version 20250324 to prevent parsing failures

- Add regex patterns for replacing LaTeX symbols \fint and \up with their Unicode equivalents

- Add \textunderscore to the list of LaTeX patterns
- This allows the model to properly render underscore characters

- Improve \left and \right command handling in LaTeX formulas
- Enhance environment type matching for array, matrix, and other structures
- Refactor code for better readability and maintainability

- Refactor LaTeX left/right pair fixing logic for better balance
- Add environment detection and correction for common math environments
- Implement more robust whitespace handling and command substitution
- Optimize regex patterns for improved performance and readability

- Add functions to fix LaTeX left and right commands
- Implement brace matching and repair in LaTeX formulas
- Remove unnecessary whitespace and repair LaTeX code
- Replace specific LaTeX commands with appropriate alternatives
- Add logging for debugging purposes

- Add functions to fix LaTeX left and right commands
- Implement brace matching and repair in LaTeX formulas
- Remove unnecessary whitespace and repair LaTeX code
- Replace specific LaTeX commands with appropriate alternatives
- Add logging for debugging purposes

- Preserve "\ " sequences during whitespace removal
- Add temporary substitution to prevent incorrect processing of "\ " sequences
- Restore "\ " sequences after removing unnecessary whitespace

- Replace get_device() function call with direct 'device' variable usage
- Simplify device configuration in OCR model initialization

- Add new Chinese OCR model (ch_PP-OCRv4_rec_server_doc_infer) for server-side use
- Update language support in app.py to include new Chinese model
- Modify models_config.yml to add new model configuration

- Automatically change to ch_lite model when using CPU for Chinese OCR
- This modification improves performance on CPU devices

- Remove OCR engine instantiation inside the loop
- Pass language directly to the table model instead of OCR engine
- Simplify code structure and improve readability

- Add text region handling in get_res_list_from_layout_res function
- Implement remove_overlaps_min_blocks function to handle overlapping blocks
- Update OCR region handling to include text regions
- Improve overlap resolution for all regions in layout results

- Update the range used to generate images_with_extra_info to match the number of images
- This fixes a potential IndexError when the number of images differs from the dataset length

- Add batch processing logs to track the progress of image analysis
- Display the current batch number, total batches, and the number of processed pages

- Update batch processing logic for improved efficiency
- Refactor image analysis and inference methods
- Optimize dataset handling and image retrieval
- Improve error handling and logging in batch processes

- Comment out the line that updates det_count in batch_analyze.py
- Add a new OCR model configuration for Chinese (ch_lite) in models_config.yml- Update the Chinese OCR model configuration to use a different recognition model

- Simplify aspect ratio calculation using direct coordinate subtraction
- Remove unnecessary list append operation
- Improve code readability and performance in table rotation detection

- Implement table orientation detection to identify if a table is in portrait mode
- Add rotation logic to turn portrait tables 90 degrees clockwise before OCR
- Update OCR processing to work with potentially rotated images
- Improve text box analysis to determine if a table is rotated

- Update predict_rec.py to check for NaN values in recognition results
- Replace NaN scores with 0.0 to ensure stability and consistency

- Add functions to calculate IoU, check if tables are inside each other, and merge tables
- Implement table merging for high IoU tables
- Add filtering to remove nested tables that don't overlap but cover a large area
- Update table_res_list and layout_res to reflect these changes

- Update OCR score formatting in batch_analyze.py and pdf_parse_union_core_v2.py
- Change score rounding method to preserve three decimal places
- Enhance accuracy representation without significantly altering the score value

- Import os and pathlib modules to handle file paths
- Define the path to the slanet-plus model
- Update RapidTableInput initialization to include the model path

- Refactor VRAM detection logic for better readability and efficiency
- Add fallback mechanism for unknown VRAM sizes
- Improve device checking in get_vram function

- Update table recognition logic to process each table individually
- Refactor layout detection to use tqdm for progress tracking
- Optimize OCR recognition by using a single tqdm wrapper
- Improve MFR prediction with a more accurate progress bar
- Simplify MFD prediction by removing unnecessary total calculation

- Remove unused imports and comments
- Increase MIN_BATCH_INFERENCE_SIZE from 100 to 200
- Comment out VRAM cleaning and logging in batch_analyze.py
- Simplify code in doc_analyze_by_custom_model.py- Add tqdm progress bar in pdf_parse_union_core_v2.py
- Enable tqdm in OCR processing

- Add tqdm progress bar to batch prediction loops in multiple model modules
- Improve logging and error handling in batch analysis script
- Update table model initialization to use default sub-model if none specified
- Add tqdm dependency to requirements.txt

- Replace ch_PP-OCRv4_det_infer.pth with ch_PP-OCRv3_det_infer.pth in models_config.yml
- Add new ch_PP-OCRv3_det_infer model configuration in arch_config.yaml

- Remove unnecessary GPU checks and cuda() calls
- Consolidate tensor device placement using .to(self.device)
- Add warning suppression for cleaner output
- Refactor conditional logic for better readability