TheFrenchOak/shotgrep-models

shotgrep models

ONNX exports of the models the shotgrep desktop app runs locally (shot detection, visual and text embeddings). The app downloads them at first launch and verifies every file by SHA-256 against its own index; nothing here is meant to be used on its own.

model	source	revision	precision	licence	size
transnet	transnetv2_pytorch/transnetv2-pytorch-weights.pth	-	fp32	MIT (TransNetV2, Souček & Lokoč)	31 MB
siglip_image	google/siglip2-so400m-patch14-384	e8e487298228002f3d8a82e0cd5c8ea9c567f57f	fp16	Apache-2.0	857 MB
siglip_text	google/siglip2-so400m-patch14-384	e8e487298228002f3d8a82e0cd5c8ea9c567f57f	fp16	Apache-2.0	1451 MB
bge_m3	BAAI/bge-m3	9a0624b896d81da7492a910ffa53731274b6cf3d	fp16	MIT	1151 MB

Licences

Each export keeps the licence of its source weights (column above). TransNetV2 weights come from the transnetv2_pytorch package.