Rigel

Rigel is a caption quality scorer for videos and images. Given a video or image, a candidate caption, and optionally reference captions, it outputs a continuous score in [0, 1].

Usage

git clone https://github.com/SHUonkei/Rigel
cd Rigel
uv sync
uv run rigel-score --media path/to/video.mp4 \
    --cand "a man is playing guitar on stage" \
    --ref "a musician performs a guitar solo"
from rigel import from_pretrained

model = from_pretrained()  # downloads this repository
scores = model(
    media_path=["path/to/video.mp4"],
    cand=["a man is playing guitar on stage"],
    refs=[["a musician performs a guitar solo"]],  # [[]] for reference-free scoring
)
Downloads last month
-
Safetensors
Model size
2.93M params
Tensor type
F32
ยท
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for ShuitsuKoyama/Rigel

Adapter
(122)
this model