| # MR-IQA-2 single-image code bundle |
|
|
| This directory contains the minimal source files needed by the released |
| single-image Actor, Editor, and Judge pipeline. It mirrors the public GitHub |
| implementation while keeping the model repository lightweight. |
|
|
| From this directory, create the two pinned inference environments and run one |
| image: |
|
|
| ```bash |
| bash scripts/setup_envs.sh --profile inference |
| python examples/quick_start.py /absolute/path/to/input.jpg |
| ``` |
|
|
| The command downloads Actor, Editor, and Judge weights from |
| `RobinY99/MR-IQA-2` as they are needed. Add `--local-files-only` when all three |
| model folders are already cached. On a host with multiple CUDA toolkits, pass |
| `--cuda-home /path/to/cuda` so runtime extensions use the intended `nvcc`. |
|
|
| The bundle includes: |
|
|
| - `examples/quick_start.py`: sequential Actor, Editor, and Judge entry point; |
| - `examples/actor_to_editor.py`: Actor generation and strict output parsing; |
| - `actor/plugin/`: public Actor output and prompt contracts; |
| - `judge/`: frozen E5 Judge contract and local inference implementation; |
| - `environment/`, `requirements/`, `scripts/setup_envs.sh`: pinned setup files. |
|
|
| Use the full [GitHub repository](https://github.com/RobinY99/MR-IQA-2) for |
| training, complete evaluation, datasets, launchers, and the full contract test |
| suite. |
|
|
