lintabsort-aidan

LinTabSort model mirror

This repo is a re-hosting mirror for the AI models used by LinTabSort, a free Windows photo-organizing app. None of these weights were trained here — this repo exists only so the app has a stable, first-party URL to download from (some original sources are personal research mirrors, community orgs, or GitHub release assets not meant for long-term production use). No original code or model architecture in this repo was authored by this account; each file below is a straight re-host (or a from-scratch ONNX re-export of the same original weights, noted where applicable) of its own upstream source. The app itself does not necessarily download from this mirror by default for every model — see each file's note.

Files and their real licenses/sources

• adain-vgg.onnx, adain-decoder.onnx — AdaIN style transfer ("Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization", Huang & Belongie, ICCV 2017). MIT licensed, from naoto0804/pytorch-AdaIN (Copyright Naoto Inoue), converted to ONNX by axinc-ai/ailia-models.
• facemesh_landmark.onnx — MediaPipe Face Mesh (Google). Apache-2.0 licensed, per google-ai-edge/mediapipe's own repository LICENSE. TFLite-to-ONNX converted by PINTO0309 (PINTO_model_zoo).
• iris_landmark.onnx — MediaPipe Iris (Google). Apache-2.0 licensed, same MediaPipe repository as above. TFLite-to-ONNX converted by PINTO0309 (PINTO_model_zoo).
• face_blendshapes.onnx — MediaPipe Face Blendshapes (Google), Apache-2.0 licensed. The original weights are distributed as a PyTorch checkpoint by py-feat/mp_blendshapes on Hugging Face (also Apache-2.0). This ONNX file is a from-scratch PyTorch→ONNX export, done for this project, using the model architecture and constants published in the py-feat package (MIT licensed, Copyright Jin Hyun Cheong, Tiankang Xie, Sophie Byrne, Eshin Jolly, Luke Chang) — no pre-converted ONNX file existed anywhere at the time this was made.
• realesr-general-x4v3.onnx — Real-ESRGAN general x4v3 (image upscaling). Apache-2.0 licensed, from xinntao/Real-ESRGAN. Straight re-host of the ONNX export published by Heliosoph/realesrgan-onnx.
• ddcolor.onnx — DDColor (image colorization). From piddnad/DDColor, Apache-2.0 licensed. Straight re-host of the ONNX export published by facefusion/models-3.0.0.
• scunet_color_real_psnr.onnx + scunet_color_real_psnr.onnx.data — SCUNet (image denoising). From cszn/SCUNet, Apache-2.0 licensed per the re-exporter's own repository. Straight re-host of the ONNX export (two-file, external-data format — both files are required together) published by Heliosoph/scunet-onnx. Note: LinTabSort runs this model CPU-only — it reproducibly crashes the DirectML (GPU) execution provider with a native access violation, confirmed across two DirectML versions and four session-option workarounds.
• depth-anything-v2-small.onnx — Depth Anything V2, Small variant (monocular depth estimation). From DepthAnything/Depth-Anything-V2, Apache-2.0 licensed. Straight re-host of the ONNX export published by onnx-community/depth-anything-v2-small. The Base variant is intentionally not mirrored here — it is CC-BY-NC-4.0 (non-commercial), unlike the Apache-2.0 Small variant, so LinTabSort continues to fetch it directly from onnx-community when a user opts into the higher-quality depth option.
• mobile_sam_image_encoder.onnx, sam_mask_decoder_single.onnx — MobileSAM (lightweight Segment Anything). From ChaoningZhang/MobileSAM, Apache-2.0 licensed. Straight re-host of the ONNX export published by Acly/MobileSAM.
• u2netp.onnx — U2-Net-P (lightweight salient object / subject segmentation). From xuebinqin/U-2-Net, Apache-2.0 licensed. Straight re-host of the ONNX export published by martintomov/comfy.
• inpainting_lama_2025jan.onnx — LaMa (image inpainting). From advimman/lama, Apache-2.0 licensed. Straight re-host of the ONNX export published by opencv/inpainting_lama (part of the OpenCV Zoo).
• face_detection_yunet_2023mar.onnx — YuNet (face detection). MIT licensed, Copyright Shiqi Yu, from the OpenCV Zoo.
• face_recognition_sface_2021dec.onnx — SFace (face recognition). Apache-2.0 licensed, from the OpenCV Zoo.
• object_detection_yolox_2022nov_int8.onnx — YOLOX, int8 quantized (object detection). Apache-2.0 licensed, from the OpenCV Zoo, itself re-hosting Megvii-BaseDetection/YOLOX.
• ediffiqa_tiny_jun2024.onnx — eDifFIQA-Tiny (face image quality assessment). CC-BY-4.0 licensed (attribution required — see LIQA/eDifFIQA), re-hosted via the OpenCV Zoo.
• disk.onnx, disk_lightglue.onnx — DISK local features + LightGlue matcher, used together for wide-baseline image matching. DISK from cvlab-epfl/disk (Apache-2.0), LightGlue from cvg/LightGlue (Apache-2.0). Straight re-host of the ONNX exports published as GitHub release assets by fabio-sim/LightGlue-ONNX (v0.1.0, also Apache-2.0). Note: this app's SuperPoint-free choice (DISK, not SuperPoint) is deliberate — SuperPoint's own upstream license is non-commercial.

Not yet mirrored here

• resnet18_places365.onnx + IO_places365.txt — a Places365 (indoor/outdoor scene classification) ResNet-18, MIT licensed per CSAILVision/places365. No pre-converted ONNX export exists publicly; this was previously converted from the original PyTorch checkpoint for this project but is not currently hosted anywhere, including here. The one feature that used it (an outdoor/indoor photo filter feeding into an in-app photo-matching tool) is currently hidden from the app's UI.

Each file's real license is the one listed above for that file, not a single blanket license for the whole repo.