Portman2 analysis models
The ONNX models Portman2, a desktop photo library, downloads for its local analysis: finding faces, finding a photograph's subject, and words for what is in a photograph. They run on the user's own computer through ONNX Runtime; nothing is uploaded anywhere.
These are byte-for-byte copies of files the original projects publish, kept here so the app's downloads do not depend on someone else's repository staying where it is. Nothing has been converted or changed, and each file's SHA-256 is the one Portman2 checks after download. The image-restoration models Portman2's Enhance… uses are in portmanphoto/portman2-enhance.
| Folder | Model | Copied from | Licence |
|---|---|---|---|
faces/ |
UltraFace RFB-320, face detection | onnx/models, from Linzaer/Ultra-Light-Fast-Generic-Face-Detector-1MB | MIT, see faces/LICENSE |
saliency/ |
U²-Net (u2netp), salient object detection | danielgatis/rembg v0.0.0 release, from xuebinqin/U-2-Net | Apache-2.0, see saliency/LICENSE; rembg's packaging MIT, saliency/LICENSE-rembg |
clip-vit-b32/ |
OpenAI CLIP ViT-B/32, image and text towers, fp32 and fp16 | Xenova/clip-vit-base-patch32 @ d15189d |
MIT, see clip-vit-b32/LICENSE |
clip-vit-b16/ |
OpenAI CLIP ViT-B/16, as above | Xenova/clip-vit-base-patch16 @ 342fdf2 |
MIT, see clip-vit-b16/LICENSE |
clip-vit-l14/ |
OpenAI CLIP ViT-L/14, as above | Xenova/clip-vit-large-patch14 @ c307790 |
MIT, see clip-vit-l14/LICENSE |
The CLIP files are Xenova's ONNX exports of openai/CLIP:
vision_model.onnx and text_model.onnx, and their _fp16 halves.
Credit for every model belongs to its authors; this repository only hosts copies under their licences.