{
  "id": "rtdetr-v2-r50vd",
  "identity": {
    "family": "rt-detr",
    "display_name": "RT-DETRv2 R50vd",
    "hub_id": "PekingU/rtdetr_v2_r50vd",
    "api_name": null,
    "revision": null,
    "variants": [
      "PekingU/rtdetr_v2_r18vd"
    ],
    "license": "Apache-2.0",
    "commercial_use": true,
    "commercial_use_note": "Apache-2.0 weights and code; no use restrictions.",
    "release_date": "2024-07",
    "paper_url": "https://arxiv.org/abs/2407.17140",
    "repo_url": "https://github.com/lyuwenyu/RT-DETR"
  },
  "task": {
    "category": "detection",
    "modalities_in": [
      "image"
    ],
    "modalities_out": [
      "boxes",
      "labels",
      "scores"
    ],
    "best_at": "RT-DETR with a better training recipe (bag of freebies); same cost, higher COCO AP.",
    "worst_at": "Closed COCO label set."
  },
  "io": {
    "input": {
      "image": {
        "resolution": "640x640",
        "color": "RGB"
      },
      "prompts": "none (fixed classes)"
    },
    "output": {
      "boxes": "xyxy pixels (post_process_object_detection)",
      "labels": "COCO 80 classes",
      "scores": "0..1"
    }
  },
  "mixpeek_wiring": {
    "extractors": [],
    "feature_uris": [],
    "vector_indexes": [],
    "retriever_stages": [],
    "composes_with": [
      "detector boxes -> crops -> SigLIP vectors per object (one document per object, or a multi-vector per frame)",
      "labels -> payload field -> attribute filter in a retriever",
      "scene splits -> keyframes -> detector -> object documents with frame timestamps"
    ]
  },
  "run": {
    "status": "plugin",
    "status_reason": "Not served. Permissive alternative to Ultralytics YOLO. Would run as a custom extractor plugin loading safetensors with transformers (trust_remote_code false), emitting boxes, labels and crops.",
    "serve_apps": [],
    "runtime": "transformers RTDetrV2ForObjectDetection + RTDetrImageProcessor",
    "weights_format": "safetensors",
    "trust_remote_code": false,
    "size_on_disk_mb": 172,
    "params_millions": 43,
    "preprocessing": "Resize to 640x640, rescale to 0..1 (image processor).",
    "postprocessing": "Sigmoid class scores, top-k queries, threshold; no NMS (one-to-one matching).",
    "serving_shape": "both",
    "batch_size": null,
    "serve_batch": null,
    "scale_to_zero": {
      "min_replicas": 0,
      "max_replicas": 1,
      "idle_timeout_s": 600,
      "memory_gb": 2,
      "num_cpus": 2,
      "num_gpus": 0,
      "cold_start_s": {
        "warm_node_s": null,
        "new_node_s": null,
        "source": null
      },
      "cpu_fallback": true
    }
  },
  "cost": {
    "unmeasured": true,
    "measured": null,
    "note": "No committed measurement; plan row 3 measures it in the standalone image."
  },
  "evidence": {
    "unmeasured": true,
    "items": []
  },
  "refs": [
    "hf://PekingU/rtdetr_v2_r50vd@282494075698cab9faa1096ae26856890030c817"
  ],
  "formats": [
    {
      "ref": "hf://PekingU/rtdetr_v2_r50vd@282494075698cab9faa1096ae26856890030c817",
      "format": "safetensors",
      "file": "model.safetensors",
      "loader": "transformers / open_clip from_pretrained; Mixpeek Hub import (api/models/huggingface.py)",
      "supported_today": true
    }
  ],
  "use_cases": [
    "object-detection"
  ],
  "recommended": {
    "value": false,
    "for": [],
    "reason": null
  },
  "demos": [],
  "seo": {
    "title": "RT-DETRv2 R50vd: object detection model",
    "description": "RT-DETRv2 R50vd is a image object detection model (Apache-2.0). Can run on Mixpeek as a custom plugin.",
    "keywords": [
      "RT-DETRv2 R50vd",
      "rt-detr",
      "object detection",
      "PekingU/rtdetr_v2_r50vd",
      "object-detection"
    ],
    "faqs": [
      {
        "q": "What is RT-DETRv2 R50vd best at?",
        "a": "RT-DETR with a better training recipe (bag of freebies); same cost, higher COCO AP."
      },
      {
        "q": "What are the limits of RT-DETRv2 R50vd?",
        "a": "Closed COCO label set."
      },
      {
        "q": "Can RT-DETRv2 R50vd be used commercially?",
        "a": "Yes. Apache-2.0 weights and code; no use restrictions."
      },
      {
        "q": "How does RT-DETRv2 R50vd run on Mixpeek?",
        "a": "Not served. Permissive alternative to Ultralytics YOLO. Would run as a custom extractor plugin loading safetensors with transformers (trust_remote_code false), emitting boxes, labels and crops."
      }
    ]
  },
  "last_verified": "2026-11-05",
  "sources": [
    "https://huggingface.co/PekingU/rtdetr_v2_r50vd",
    "https://arxiv.org/abs/2407.17140",
    "https://github.com/lyuwenyu/RT-DETR"
  ],
  "registry": null,
  "templates": [],
  "usage": {
    "namespaces": null,
    "collections": null,
    "uncounted": true,
    "templates": 0,
    "counted_at": null,
    "source": null
  },
  "facets": {
    "status": "plugin",
    "category": "detection",
    "family": "rt-detr",
    "license": "Apache-2.0",
    "commercial_use": true,
    "modalities_in": [
      "image"
    ],
    "modalities_out": [
      "boxes",
      "labels",
      "scores"
    ],
    "use_cases": [
      "object-detection"
    ],
    "recommended": false,
    "release_date": "2024-07",
    "params_millions": 43,
    "size_on_disk_mb": 172,
    "vector_dims": null,
    "vector_distance": null,
    "weights_format": "safetensors",
    "formats": [
      "safetensors"
    ],
    "loadable_today": true,
    "trust_remote_code": false,
    "cpu_fallback": true,
    "num_gpus": 0,
    "memory_gb": 2,
    "min_replicas": 0,
    "serving_shape": "both",
    "served": false,
    "ref_schemes": [
      "hf"
    ],
    "has_evidence": false,
    "cost_measured": false,
    "template_count": 0,
    "usage_namespaces": null,
    "usage_collections": null,
    "usage_uncounted": true,
    "has_demo": false,
    "rank_score": 1
  },
  "rank_score": 1
}