ONNX models mirrored for PhotoPrism (https://photoprism.app/). Each model below is redistributed under the license stated by its upstream publisher. This file records attribution only and grants no rights of its own. A SHA-256 checksum is published next to every model as .sha256, and PhotoPrism verifies it on download. Models mirrored only as benchmark comparators are not listed here. They are kept under testing/ with their own NOTICE, and are not intended for production use. Some of these files are exported by PhotoPrism rather than copied from the publisher, because the publisher offers no ONNX artifact. Those entries name the source checkpoint and its checksum, so the chain can be followed back to the publisher's own file; the model's own checksum identifies our export and matches nothing upstream. Every source checkpoint recorded here has been verified against the publisher by SHA-256. Training data is recorded per model. Where a dataset attaches terms to its own distribution, those terms bind whoever obtained the dataset; they are noted here as part of the provenance record. -------------------------------------------------------------------------------- face_recognition_sface_2021dec.onnx SFace face recognition model. Upstream: OpenCV Zoo https://github.com/opencv/opencv_zoo/tree/main/models/face_recognition_sface Reference: Zhong et al., "SFace: Sigmoid-Constrained Hypersphere Loss for Robust Face Recognition" - https://ieeexplore.ieee.org/document/9318547 https://github.com/zhongyy/SFace License: Apache License 2.0 https://www.apache.org/licenses/LICENSE-2.0 The upstream directory states: "All files in this directory are licensed under Apache 2.0 License." Redistributed unmodified. -------------------------------------------------------------------------------- auraface_v1_glintr100.onnx AuraFace v1 face recognition model (ArcFace-architecture, independently trained on commercially and publicly available data). Upstream: fal - https://huggingface.co/fal/AuraFace-v1 Published as glintr100.onnx; renamed here to avoid collision with the identically named model in InsightFace's antelopev2 pack, which is a different file under different terms. Reference: Deng et al., "ArcFace: Additive Angular Margin Loss for Deep Face Recognition" - https://arxiv.org/abs/1801.07698 License: Apache License 2.0 https://www.apache.org/licenses/LICENSE-2.0 Redistributed unmodified. -------------------------------------------------------------------------------- face_detection_yunet_2026may.onnx YuNet face detection model (bounding boxes + 5 keypoints). Upstream: OpenCV Zoo https://github.com/opencv/opencv_zoo/tree/main/models/face_detection_yunet Trained by: https://github.com/ShiqiYu/libfacedetection.train (BSD-3-Clause) Reference: Wu et al., "YuNet: A Tiny Millisecond-level Face Detector" https://link.springer.com/article/10.1007/s11633-023-1423-y License: MIT License, Copyright (c) 2020 Shiqi Yu The upstream directory states: "All files in this directory are licensed under MIT License." Redistributed unmodified. This export uses dynamic input dimensions. -------------------------------------------------------------------------------- efficientformerv2_s1.onnx EfficientFormerV2-S1 image classification model (ImageNet-1k, 1000 classes). Upstream: Weights published by timm (Ross Wightman, Hugging Face) https://huggingface.co/timm/efficientformerv2_s1.snap_dist_in1k Architecture and original checkpoints by Snap Research https://github.com/snap-research/EfficientFormer Reference: Li et al., "Rethinking Vision Transformers for MobileNet Size and Speed" - https://arxiv.org/abs/2212.08059 License: Apache License 2.0, as stated by the timm model card. https://www.apache.org/licenses/LICENSE-2.0 The originating Snap Research repository is published under its own permissive license, which carries third-party notices. Training data: ImageNet-1k (ILSVRC-2012-CLS). Source: model.safetensors at revision 0c1fc60a0e89b6309d9de451cdacc11ac0a8b987 SHA256 3ef67aa25321c2a3d145e12ef4823e5f7da4356ed1656f9f1bd6b9af339e5655 Exported to ONNX by PhotoPrism from the checkpoint above. -------------------------------------------------------------------------------- efficientformerv2_s2.onnx EfficientFormerV2-S2 image classification model (ImageNet-1k, 1000 classes). Upstream: Weights published by timm (Ross Wightman, Hugging Face) https://huggingface.co/timm/efficientformerv2_s2.snap_dist_in1k Architecture and original checkpoints by Snap Research https://github.com/snap-research/EfficientFormer Reference: Li et al., "Rethinking Vision Transformers for MobileNet Size and Speed" - https://arxiv.org/abs/2212.08059 License: Apache License 2.0, as stated by the timm model card. https://www.apache.org/licenses/LICENSE-2.0 The originating Snap Research repository is published under its own permissive license, which carries third-party notices. Training data: ImageNet-1k (ILSVRC-2012-CLS). Source: model.safetensors at revision 1c56a76355000c79568559d34ba3fa24416b5107 SHA256 c46c28768173ee0e518b7cd94af1e832cb88d2115068a81386c08650cc22a24f Exported to ONNX by PhotoPrism from the checkpoint above. -------------------------------------------------------------------------------- repvit_m1_0.onnx RepViT-M1.0 image classification model (ImageNet-1k, 1000 classes). Upstream: Weights published by timm (Ross Wightman, Hugging Face) https://huggingface.co/timm/repvit_m1_0.dist_300e_in1k Architecture and original checkpoints by THU-MIG https://github.com/THU-MIG/RepViT Reference: Wang et al., "RepViT: Revisiting Mobile CNN From ViT Perspective" https://arxiv.org/abs/2307.09283 License: Apache License 2.0, as stated by the timm model card and by the THU-MIG repository. https://www.apache.org/licenses/LICENSE-2.0 Training data: ImageNet-1k (ILSVRC-2012-CLS). Source: model.safetensors at revision 94445f5481b027599200e61ed5e108dbaedc0139 SHA256 792cf8a2e2651a1584bb5f23b9e38c3bfdf38da0dd2ecc6f0bdffa858356daa0 Exported to ONNX by PhotoPrism from the checkpoint above, with the model reparameterized for inference. -------------------------------------------------------------------------------- efficientnet_b0.onnx EfficientNet-B0 image classification model (ImageNet-1k, 1000 classes), RandAugment training recipe. Upstream: Weights trained and published by timm (Ross Wightman, Hugging Face) https://huggingface.co/timm/efficientnet_b0.ra_in1k https://github.com/huggingface/pytorch-image-models Reference: Tan and Le, "EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks" - https://arxiv.org/abs/1905.11946 License: Apache License 2.0 https://www.apache.org/licenses/LICENSE-2.0 Training data: ImageNet-1k (ILSVRC-2012-CLS). Source: model.safetensors at revision 1b5383e5f79cc0f7fc067e372f8f26a5fa73f26a SHA256 d569899762ea9b1384ee07f4af64805cf8caa1c55f9253ebb1080dc40e87a2cd Exported to ONNX by PhotoPrism from the checkpoint above. -------------------------------------------------------------------------------- yahoo_open_nsfw.onnx Yahoo OpenNSFW detection model (thin ResNet-50 1by2, two classes). The publisher documents its scope as pornographic imagery, and excludes sketches, cartoons, text and graphic violence. Upstream: Yahoo Inc. - https://github.com/yahoo/open_nsfw License: BSD 2-Clause License, Copyright (c) 2016, Yahoo Inc. https://github.com/yahoo/open_nsfw/blob/master/LICENSE.md The Caffe weights are published inside that repository rather than under separately stated terms. Training data: Not disclosed by the publisher. Source: nsfw_model/resnet_50_1by2_nsfw.caffemodel at revision a4e13931465f4380742545932657eeea0a10aa48 SHA256 74b7c0374425c2b324f73b134f8a7254eab9bbddc9e452a22d934b8c17015ad2 nsfw_model/deploy.prototxt, same revision SHA256 a359e2d7b23abca938bb52580acf578c1b78870263fea814afa294fb4f315a4b Converted from Caffe to ONNX by PhotoPrism from the files above. -------------------------------------------------------------------------------- falconsai_nsfw_image_detection_224.onnx Falconsai NSFW image detection model (Vision Transformer Base, patch 16, two classes). Upstream: Falconsai https://huggingface.co/Falconsai/nsfw_image_detection License: Apache License 2.0, as stated by the model card. https://www.apache.org/licenses/LICENSE-2.0 Training data: Proprietary; exact sources not disclosed by the publisher. Source: model.safetensors at revision 04367978d3474804ab1a00a9bd6548b741764069 SHA256 97b2ce64ec146884b37f98ee7944ca4891aa72f6827dc0cb10684a1cbecd5830 Exported to ONNX by PhotoPrism from the checkpoint above. -------------------------------------------------------------------------------- freepik_nsfw_image_detector.onnx Freepik NSFW image detector (EVA02 Base, patch 14, four severity classes). Upstream: Freepik https://huggingface.co/Freepik/nsfw_image_detector License: MIT License, as stated by the model card. Training data: Synthetic labels; exact source images not disclosed by the publisher. Source: model.safetensors at revision 15b85477e4fd2000db76ae9aae0f89a72f95e2e3 SHA256 024a9d4818fae2656403bf626c9f8c9e7789c2da274749fbebb1060d8fdaa7ab Exported to ONNX by PhotoPrism from the checkpoint above. -------------------------------------------------------------------------------- adamcodd_vit_base_nsfw_fp32.onnx adamcodd_vit_base_nsfw_int8.onnx AdamCodd ViT-Base NSFW detector (Vision Transformer Base, patch 16, two classes), in full and quantized precision. Upstream: AdamCodd https://huggingface.co/AdamCodd/vit-base-nsfw-detector Published as onnx/model.onnx and onnx/model_int8.onnx at revision 8587de998f441aac03fdd57a85d2e4cb808c7d64; renamed here to record the publisher and the precision in the filename. License: Apache License 2.0, as stated by the model card. https://www.apache.org/licenses/LICENSE-2.0 Training data: Partially described by the publisher; exact sources are not disclosed. Redistributed unmodified, and verified byte-identical to the publisher's files by SHA-256.