VigIA uses multiple ML models from different sources (Google Drive, PyTorch Hub, insightface). Many of these sources are unreliable for automated downloads (captchas, rate limits, broken URLs). This repo consolidates all weights in one CDN-backed location for reliable deployment.

Models Included

Subfolder	Model	Task	Original Source	License
`osnet-ain/`	OSNet-AIN x1.0	Person Re-ID	torchreid (Google Drive)	MIT
`solider-swin-s/`	SOLIDER Swin-S	Person Re-ID backbone	tinyvision (Google Drive)	Apache 2.0
`solider-swin-b/`	SOLIDER Swin-B	Person Re-ID backbone	tinyvision (Google Drive)	Apache 2.0
`transreid-vehicle/`	TransReID ViT	Vehicle Re-ID	damo-cv (Google Drive)	MIT
`resnet50-ibn/`	ResNet50-IBN	Vehicle Re-ID backbone	torchvision (PyTorch Hub)	BSD-3
`scrfd-2.5g/`	SCRFD-2.5GF	Face Detection	insightface (buffalo_l.zip)	MIT
`scrfd-10g/`	SCRFD-10GF	Face Detection	insightface (buffalo_l.zip)	MIT
`scrfd-34g/`	SCRFD-34GF	Face Detection	insightface (OneDrive)	MIT

Usage

python

from huggingface_hub import snapshot_download

# Download a specific model subfolder
snapshot_download(
    repo_id='rotivrotiv/model-weights',
    allow_patterns='solider-swin-b/**',
    local_dir='./models/',
)

License

Each model retains its original license (see table above). All models are commercially compatible (Apache 2.0, MIT, BSD-3). No model in this repo uses GPL, AGPL, or non-commercial licenses.

Credits

SOLIDER — tinyvision, ICCV 2023
TransReID — DAMO Academy, ICCV 2021
OSNet — torchreid, MIT
SCRFD — insightface, MIT
ResNet50 — torchvision, BSD-3