NATSU-FlorenceCaptioner/README.md

38 lines
804 B
Markdown

# Florence-2 Captioning Pipeline
High-throughput asynchronous captioning pipeline using **Florence-2 Base PromptGen**.
## Goals
- Download images from S3/HTTP concurrently
- Preprocess (resize/normalize)
- Run batched caption generation on GPU
- Persist captions back to a database (async)
## Project structure
- `src/`: implementation code
- `tests/`: unit/integration tests
- `todo.md`: tasks list
- `implementationPlanV2.md`: architecture + design notes
## Quickstart
1. Install dependencies:
```bash
pip install -r requirements.txt
```
2. Configure environment variables (see `src/config.py` for expected vars).
3. Run the pipeline (example):
```bash
python -m src.pipeline --dry-run
```
## Notes
This repo is intended as a foundation for building a fast, async dataset captioning tool.