flowchart LR
    A[Code Changes] --> B[Remove Empty Line]
    A --> C[Add Diagram Examples]
    C --> D[Visualize Workflows]
``` |
| source/model/etl/code/postprocess/nms.py | 111 added, 0 removed | This code implements Non-Maximum Suppression (NMS) algorithms for single-class and multi-class object detection, along with post-processing functions for model outputs. |
| source/model/etl/code/sm_predictor.py | 23 added, 12 removed | The code changes involve removing the import of `lambda_return` function, adding a `create_response` function to handle Flask responses with CORS headers, modifying the `handler` function to use `create_response` instead of `lambda_return`, and updating the `/transformation` route to use the new response handling approach. |
| source/model/etl/code/layout.py | 17 added, 15 removed | The code changes involve restructuring the imports, introducing configuration files for GPU provider and model settings, and modifying the LayoutPredictor class to use these configurations. Additionally, it adds aspect ratio handling and configurable thresholds for non-maximum suppression and score filtering. |
| source/model/etl/code/utils.py | 107 added, 151 removed | The code changes involve replacing the previous object detection code with new functions for reading images from various sources (URLs, S3, base64), handling different image formats (GIF, PDF), and creating a standardized Lambda function response structure. |
| source/model/etl/code/main.py | 39 added, 10 removed | The code changes introduce concurrent processing of figures using the ThreadPoolExecutor from the concurrent.futures module. It replaces the sequential processing of figures with a parallel approach, where each figure is processed concurrently using a separate thread. This change aims to improve performance by utilizing multiple CPU cores for figure understanding tasks. |
| source/model/etl/code/figure_llm.py | 110 added, 35 removed | The code changes introduce support for using OpenAI models in addition to Bedrock models, add logging, and refactor the codebase to improve modularity and maintainability. The key changes include handling API keys from AWS Secrets Manager, invoking OpenAI models via the OpenAI API, and separating prompts into individual files. |
| source/model/etl/code/table.py | 21 added, 17 removed | The code changes introduce a configuration file (model_config.py) to store model-related settings like session options, preprocessing steps, postprocessing parameters, and table matching configurations. The session options, preprocessing operators, and postprocessing parameters are now loaded from this configuration file instead of being hardcoded. Additionally, file paths are constructed using os.path.join for better cross-platform compatibility. |
| source/model/etl/code/ocr.py | 16 added, 43 removed | The code changes involve the following:

1. Introduced a `get_provider_config` function to determine the execution provider and batch size based on GPU availability.
2. Used `os.path.join` for constructing file paths.
3. Moved model configuration details to a separate `MODEL_CONFIGS` module.
4. Simplified model path construction based on language using `MODEL_CONFIGS`.
5. Consolidated post-processing parameter configuration using `MODEL_CONFIGS`. |

</details>
  

</details>



<details>
<summary>🤖 AI-Generated PR Description (Powered by Amazon Bedrock)</summary>

# Description
This pull request introduces several changes to the project, including the addition of new files, removal of existing files, and modifications to existing files. The primary changes are as follows:

- Added new files for configuration management (`config.py`, `gpu_config.py`, `model_config.py`), image augmentation (`imaug/preprocess.py`), non-maximum suppression (`postprocess/nms.py`), and prompt templates (`prompt/chart.txt`, `prompt/description.txt`).
- Removed the `aikits_utils.py` file.
- Modified existing files for figure processing (`figure_llm.py`), image augmentation (`imaug/__init__.py`), layout management (`layout.py`), main execution (`main.py`), optical character recognition (`ocr.py`), post-processing (`postprocess/__init__.py`), mermaid template (`prompt/mermaid_template.txt`), requirements (`requirements.txt`), prediction (`sm_predictor.py`), table processing (`table.py`), and utilities (`utils.py`).
- Encountered an error with the `untitled.txt` file.

These changes aim to enhance the project's functionality, improve configuration management, introduce image augmentation techniques, implement non-maximum suppression for post-processing, and provide additional prompt templates for figure and description generation.

## Type of change
- [ ] Bug fix (non-breaking change which fixes an issue)
- [x] New feature (non-breaking change which adds functionality)
- [x] Breaking change (fix or feature that would cause existing functionality to not work as expected)
- [x] This change requires a documentation update
## File Stats Summary

File number involved in this PR: *19*, unfold to see the details:

<details>

The file changes summary is as follows:

| <div style="width:150px">Files</div> | <div style="width:160px">Changes</div> | <div style="width:320px">Change Summary</div> |
|:-------|:--------|:--------------|
| source/model/etl/code/aikits_utils.py | 0 added, 50 removed | This file is removed in this PR |
| source/model/etl/code/imaug/preprocess.py | 35 added, 0 removed | This code defines a function `preprocess` that preprocesses an input image for a model by resizing and padding it to a specified input size, and swapping channels if needed. |
| source/model/etl/code/imaug/__init__.py | 4 added, 0 removed | The code changes import the `preprocess` module and add it to the `__all__` list for exposing it publicly. |
| source/model/etl/code/postprocess/__init__.py | 7 added, 1 removed | The code changes introduce imports for nms, multiclass_nms, and postprocess functions, and update the __all__ list to include these functions. |
| source/model/etl/code/requirements.txt | 2 added, 1 removed | The code changes involve adding the 'openai' library version 1.0.0 or higher to the list of required dependencies. |
| source/model/etl/code/config.py | 73 added, 0 removed | This code defines a ModelConfig class that manages configurations for different language models (Chinese and English) for text detection and recognition tasks. It provides methods to retrieve model paths, dictionary paths, and post-processing configurations based on the specified language and model type. |
| source/model/etl/code/prompt/mermaid_template.txt | 0 added, 1 removed | The code changes remove an empty line from the documentation, making the content more concise.

```mermaid
graph LR
A[Code Changes] --> B[Remove Empty Line]
B --> C[Concise Documentation]
``` |
| source/model/etl/code/layout.py | 17 added, 15 removed | The code changes involve restructuring the imports, introducing configuration files for model and GPU settings, and adding configuration parameters for non-maximum suppression threshold, score threshold, image size, and aspect ratio threshold to the LayoutPredictor class. |
| source/model/etl/code/main.py | 39 added, 10 removed | The code changes introduce concurrent processing of figures using ThreadPoolExecutor for figure understanding, with a configurable maximum number of workers. It replaces the sequential processing of figures with concurrent processing, potentially improving performance. Additionally, it logs the time taken for figure processing. |
| source/model/etl/code/prompt/description.txt | 8 added, 0 removed | This code change introduces a task for an experienced image analysis expert to describe the details shown in a given illustration, including any text present, while utilizing the provided context enclosed within <doc></doc> tags to better understand and describe the image. The expert's description should be written within <output></output> XML tags. |
| source/model/etl/code/gpu_config.py | 13 added, 0 removed | The code checks for available GPUs, sets execution providers, recognition batch size, and layout model path accordingly for CPU or GPU execution. |
| source/model/etl/code/model_config.py | 78 added, 0 removed | This code defines configuration dictionaries for a model, layout analysis, and table recognition in an OCR (Optical Character Recognition) system, with settings for different languages, preprocessing/postprocessing steps, and model paths. |
| source/model/etl/code/figure_llm.py | 110 added, 35 removed | This code update adds support for using OpenAI models alongside Anthropic's Bedrock models, retrieves API keys from AWS Secrets Manager, and includes new prompts for chart descriptions and image descriptions. It also adds logging configuration and handles environment variables for OpenAI API base URL. |
| source/model/etl/code/sm_predictor.py | 23 added, 12 removed | The code changes involve refactoring the handler function to create a separate create_response function for generating Flask responses with CORS headers. The lambda_return import is removed, and the handler function now returns the response directly using create_response. The transformation function is also updated to use the new create_response function. |
| source/model/etl/code/utils.py | 107 added, 151 removed | The code changes involve refactoring to add functions for reading images from various sources (URLs, S3, base64) and handling different image formats (GIF, PDF). It also adds a utility function for creating a standardized Lambda function response. |
| source/model/etl/code/postprocess/nms.py | 111 added, 0 removed | The code changes implement non-maximum suppression (NMS) algorithms for single and multi-class object detection, along with post-processing of model outputs for object detection. |
| source/model/etl/code/ocr.py | 16 added, 43 removed | The code changes involve refactoring and modularization of the code. It introduces new modules `gpu_config` and `model_config` to handle GPU configuration and model configurations respectively. The initialization of TextDetector and TextRecognizer classes has been simplified by using configurations from `model_config`. The file paths are now constructed using `os.path.join` for better cross-platform compatibility. Additionally, some redundant code and comments have been removed. |
| source/model/etl/code/prompt/chart.txt | 16 added, 0 removed | This code provides instructions for converting a chart image into a Markdown table based on context information. The key steps involve finding the chart, observing its structure and data, using context to understand it, converting the data into a Markdown table format following specific guidelines, and returning only the Markdown table within XML tags. |
| source/model/etl/code/table.py | 21 added, 17 removed | The code changes introduce a configuration file (model_config.py) to store model settings like session options, preprocessing, postprocessing, and table matching parameters. The session options and postprocessing parameters are loaded from the configuration file instead of being hardcoded. The table matching configuration is also loaded from the config file. |

</details>
  

</details>