{"reference_analysis":{"checked_at":"2026-10-08","selection_basis":{"en":"Ten evaluation entry points, ordered by integration task; not an official or popularity ranking.","zh-TW":"依整合任務整理的十個評估入口；不是官方排名或人氣排行。"},"items":[{"id":"http-client","name":"Inference SDK","kind":"Python HTTP client","url":"https://docs.roboflow.com/reference/inference/inference-sdk","audience":{"en":"Backend / application developer","zh-TW":"後端／應用工程師"},"capability":{"en":"Call deployed models using one Python client across cloud and self-hosted endpoints.","zh-TW":"使用同一個 Python client 呼叫雲端或自架模型。"},"io":{"en":"Image + model identifier → prediction dictionaries","zh-TW":"影像＋模型識別碼 → 預測字典"},"runtime":{"en":"Client application; inference runs on the selected server.","zh-TW":"Client 在應用端執行；模型由指定伺服器執行。"},"constraint":{"en":"The SDK is a client, not a model runtime. Server versions affect authentication support.","zh-TW":"SDK 是呼叫端，不是模型執行環境；認證方式受伺服器版本影響。"},"validation":{"en":"Compare one fixed image on both endpoints; record errors and cold versus warm latency.","zh-TW":"同一張影像測試兩種端點，記錄錯誤與冷啟動／暖機延遲。"}},{"id":"native","name":"Inference Python","kind":"Native Python runtime","url":"https://docs.roboflow.com/reference/inference/inference-python","audience":{"en":"ML / edge engineer","zh-TW":"ML／邊緣運算工程師"},"capability":{"en":"Run model inference inside a Python process.","zh-TW":"在 Python 程序內直接執行模型。"},"io":{"en":"Image + supported model → predictions","zh-TW":"影像＋支援模型 → 預測"},"runtime":{"en":"Local CPU / GPU, subject to the model backend.","zh-TW":"本機 CPU／GPU，依模型後端支援情況而定。"},"constraint":{"en":"Model weights, runtime dependencies and hardware compatibility need separate checks.","zh-TW":"需分別確認權重、執行環境依賴與硬體相容性。"},"validation":{"en":"Pin the model and dependencies; measure memory and latency on the actual target device.","zh-TW":"固定模型與依賴版本，在目標裝置測量記憶體與延遲。"}},{"id":"cloud","name":"Serverless Cloud API","kind":"Managed inference API","url":"https://docs.roboflow.com/deployment/roboflow-cloud/serverless-api","audience":{"en":"Backend / prototype team","zh-TW":"後端／原型團隊"},"capability":{"en":"Use managed cloud inference for models and Workflows.","zh-TW":"使用託管雲端執行模型與 Workflows。"},"io":{"en":"Image / workflow inputs → inference results","zh-TW":"影像／工作流程輸入 → 推論結果"},"runtime":{"en":"Roboflow cloud; network connectivity required.","zh-TW":"Roboflow 雲端，需要網路連線。"},"constraint":{"en":"Check supported models, payload limits, credentials and current billing before estimating cost.","zh-TW":"估算成本前確認支援模型、傳輸限制、憑證與現行計費。"},"validation":{"en":"Test a representative image batch and a timeout; record total response time and usage.","zh-TW":"測試代表性影像批次與逾時情境，記錄完整回應時間與用量。"}},{"id":"server","name":"Inference Server","kind":"Self-hosted HTTP service","url":"https://docs.roboflow.com/deployment/self-hosted/inference-server","audience":{"en":"Platform / factory IT","zh-TW":"平台／工廠 IT"},"capability":{"en":"Serve models and Workflows from your chosen infrastructure.","zh-TW":"在自選基礎設施提供模型與 Workflows 服務。"},"io":{"en":"HTTP image or video request → results","zh-TW":"HTTP 影像或視訊請求 → 結果"},"runtime":{"en":"Docker service on compatible local, edge or cloud hardware.","zh-TW":"在相容本機、邊緣或雲端硬體執行 Docker 服務。"},"constraint":{"en":"Self-hosting alone does not establish offline readiness or support for every model.","zh-TW":"自架不等於已具備離線能力，也不代表支援所有模型。"},"validation":{"en":"Verify startup, weight availability, concurrent requests and restart recovery on the target host.","zh-TW":"在目標主機確認啟動、權重取得、併發請求與重啟復原。"}},{"id":"workflows","name":"Workflows SDK execution","kind":"Workflow orchestration","url":"https://docs.roboflow.com/reference/inference/inference-sdk/workflows","audience":{"en":"Solution engineer / technical PM","zh-TW":"解決方案工程師／技術 PM"},"capability":{"en":"Execute a saved Workflow or an inline specification through the SDK.","zh-TW":"透過 SDK 執行已儲存工作流程或內嵌規格。"},"io":{"en":"Images + parameters → configured workflow outputs","zh-TW":"影像＋參數 → 工作流程定義的輸出"},"runtime":{"en":"Inference endpoint selected by the client.","zh-TW":"由 client 指定的推論端點。"},"constraint":{"en":"Saved definitions can be cached; verify which revision actually executed.","zh-TW":"已儲存定義可能被快取；需確認實際執行版本。"},"validation":{"en":"Use a known input and expected output contract; change one step and verify the revised result.","zh-TW":"準備已知輸入與預期輸出契約，修改一個步驟後驗證新結果。"}},{"id":"stream","name":"WebRTC Streaming","kind":"Real-time video interface","url":"https://docs.roboflow.com/reference/inference/inference-sdk/webrtc","audience":{"en":"Video / edge application team","zh-TW":"視訊／邊緣應用團隊"},"capability":{"en":"Stream video into inference and receive frames or prediction data.","zh-TW":"串流視訊進行推論並接收畫面或預測資料。"},"io":{"en":"Camera / RTSP / file frames → timestamped results","zh-TW":"攝影機／RTSP／檔案影格 → 附時間資訊結果"},"runtime":{"en":"Client, network and inference server form the pipeline.","zh-TW":"由 client、網路與推論伺服器共同形成管線。"},"constraint":{"en":"Real-time operation may drop frames. Camera reachability and NAT behavior matter.","zh-TW":"即時模式可能捨棄影格；需處理攝影機可達性與 NAT。"},"validation":{"en":"Measure end-to-end delay and frame loss; test reconnection and missing predictions.","zh-TW":"量測端到端延遲與掉格，測試重連與預測缺失。"}},{"id":"platform-sdk","name":"Platform Python SDK","kind":"Dataset / training SDK","url":"https://docs.roboflow.com/reference/platform/python-sdk","audience":{"en":"Data / ML operations team","zh-TW":"資料／ML 維運團隊"},"capability":{"en":"Manage projects, datasets, versions and training through the roboflow package.","zh-TW":"透過 roboflow 套件管理專案、資料集、版本與訓練。"},"io":{"en":"Images / project operations → versioned data and training operations","zh-TW":"影像／專案操作 → 版本化資料與訓練操作"},"runtime":{"en":"Python application connected to the platform.","zh-TW":"連接平台的 Python 應用。"},"constraint":{"en":"Inference in this SDK is deprecated; use the dedicated inference path.","zh-TW":"此 SDK 的推論功能已棄用，應使用專用推論接口。"},"validation":{"en":"Upload a small labeled sample, export its version and compare counts and label mappings.","zh-TW":"上傳小型標註樣本、匯出指定版本，比對數量與標籤對照。"}},{"id":"rest","name":"Platform REST API","kind":"Platform management API","url":"https://docs.roboflow.com/reference/platform/rest-api","audience":{"en":"Backend / non-Python teams","zh-TW":"後端／非 Python 團隊"},"capability":{"en":"Automate workspace, project and dataset management over HTTP.","zh-TW":"透過 HTTP 自動化工作區、專案與資料集管理。"},"io":{"en":"Authenticated resource operations → JSON responses","zh-TW":"經認證資源操作 → JSON 回應"},"runtime":{"en":"Your service calls the Roboflow platform.","zh-TW":"由自有服務呼叫 Roboflow 平台。"},"constraint":{"en":"Platform management and inference use different endpoints; scope credentials to the workspace.","zh-TW":"平台管理與推論使用不同端點；憑證須符合工作區範圍。"},"validation":{"en":"Verify access with a read operation; test unauthorized requests and error handling before writes.","zh-TW":"先以讀取操作確認權限，寫入前測試未授權請求與錯誤處理。"}},{"id":"supervision","name":"Supervision","kind":"Prediction processing library","url":"https://supervision.roboflow.com/latest/","audience":{"en":"Computer-vision application team","zh-TW":"電腦視覺應用團隊"},"capability":{"en":"Convert detections into annotations, tracking, zones and counts.","zh-TW":"將偵測結果轉成標註、追蹤、區域與計數。"},"io":{"en":"Model detections + frames → processed detections / visual overlays","zh-TW":"模型偵測＋影格 → 處理後結果／視覺疊圖"},"runtime":{"en":"Python application alongside a model pipeline.","zh-TW":"與模型管線搭配的 Python 應用。"},"constraint":{"en":"This processes predictions; it is not a hosted inference service.","zh-TW":"此工具處理預測結果，本身不是託管推論服務。"},"validation":{"en":"Check coordinate conversion and count a hand-labeled clip with occlusion and re-entry.","zh-TW":"核對座標轉換，以含遮擋與重新進入情境的人工標記影片驗證計數。"}},{"id":"rfdetr","name":"RF-DETR","kind":"Model training / inference library","url":"https://rfdetr.roboflow.com/latest/","audience":{"en":"ML researcher / model owner","zh-TW":"ML 研究者／模型負責人"},"capability":{"en":"Evaluate or fine-tune transformer-based detection models for your data.","zh-TW":"評估或微調 Transformer 偵測模型以適配自有資料。"},"io":{"en":"Labeled dataset / images → model artifacts / detections","zh-TW":"標註資料集／影像 → 模型產物／偵測結果"},"runtime":{"en":"Training and inference environment matched to the model.","zh-TW":"符合模型需求的訓練與推論環境。"},"constraint":{"en":"Check the exact model variant, license and export path; published benchmarks are not device guarantees.","zh-TW":"確認確切模型版本、授權與匯出路徑；公開效能不代表目標裝置保證。"},"validation":{"en":"Use a held-out production-like set; compare missed defects, false alarms and target-device latency.","zh-TW":"使用獨立且近似產線的測試集，比較漏檢、誤報與目標裝置延遲。"}}]},"content_coverage":{"reference_guide":true,"developer_starter":true,"integration_test":"not-run","updated_at":"2026-10-08"},"developer_starter":{"version":"1.0.0","verification":"offline-fixtures-only;live-provider-not-run","filename":"roboflow_smoke.py","code":"\"\"\"Smart Tools starter: one Roboflow object-detection image. Not production-certified.\"\"\"\nimport json\nimport math\nimport os\nimport sys\nfrom datetime import datetime, timezone\nfrom importlib.metadata import version\nfrom pathlib import Path\nfrom time import monotonic\n\n\ndef normalize(raw):\n    if not isinstance(raw, dict) or not isinstance(raw.get('predictions'), list):\n        raise ValueError('unexpected_schema')\n    items = []\n    for item in raw['predictions']:\n        if not isinstance(item, dict) or not isinstance(item.get('class'), str):\n            raise ValueError('unexpected_schema')\n        score = item.get('confidence')\n        if type(score) not in (int, float) or not math.isfinite(score) or not 0 <= score <= 1:\n            raise ValueError('unexpected_schema')\n        box = {key: item.get(key) for key in ('x', 'y', 'width', 'height')}\n        if any(type(n) not in (int, float) or not math.isfinite(n) for n in box.values()):\n            raise ValueError('unexpected_schema')\n        if box['width'] < 0 or box['height'] < 0:\n            raise ValueError('unexpected_schema')\n        items.append({'label': item['class'], 'score': score, 'box_center_pixels': box})\n    return items\n\n\ndef run():\n    start = monotonic()\n    result = {'schema_version': '1.0.0', 'provider': 'Roboflow', 'status': 'error',\n              'model_version': None, 'data': None, 'error': None,\n              'captured_at': datetime.now(timezone.utc).isoformat()}\n    try:\n        key, model = os.environ['ROBOFLOW_API_KEY'], os.environ['ROBOFLOW_MODEL_ID']\n        filename = Path(os.environ['IMAGE_PATH'])\n        if not key or '/' not in model or not filename.is_file():\n            raise ValueError('configuration')\n        from inference_sdk import InferenceHTTPClient, InferenceConfiguration\n        result['sdk_version'] = version('inference-sdk')\n        result['model_version'] = model\n        # Hosted endpoint only; image leaves this machine. Header auth needs server >=1.5.\n        client = InferenceHTTPClient(api_url='https://serverless.roboflow.com', api_key=key)\n        client.configure(InferenceConfiguration(api_key_transport='header'))\n        result['data'] = normalize(client.infer(str(filename), model_id=model))\n        result['status'] = 'ok'\n    except (KeyError, FileNotFoundError):\n        result['error'] = 'configuration'\n    except ValueError:\n        result['error'] = 'configuration_or_schema'\n    except ImportError:\n        result['error'] = 'dependency_missing'\n    except Exception:\n        # Provider exceptions may contain request URLs or credentials: never echo them.\n        result['error'] = 'provider_or_transport_failure'\n    result['elapsed_ms'] = round((monotonic() - start) * 1000)\n    return result\n\n\nif __name__ == '__main__':\n    output = run()\n    print(json.dumps(output, ensure_ascii=False, allow_nan=False))\n    sys.exit(0 if output['status'] == 'ok' else 1)\n","scope":{"en":"Single-image object detection; classification, segmentation and video need separate adapters.","zh-TW":"單張影像物件偵測；分類、分割與視訊需另設轉接器。"},"setup":{"en":"Create a Python environment, install inference-sdk, and record its exact installed version. Set ROBOFLOW_API_KEY, ROBOFLOW_MODEL_ID (project/version), and IMAGE_PATH in your local environment. Save the code as roboflow_smoke.py and run python roboflow_smoke.py. The SDK call has no explicit total deadline in this starter: use a supervised process timeout for the pilot.","zh-TW":"建立 Python 環境、安裝 inference-sdk 並記錄實際版本。在本地環境設定 ROBOFLOW_API_KEY、ROBOFLOW_MODEL_ID（project/version）及 IMAGE_PATH。另存程式為 roboflow_smoke.py，執行 python roboflow_smoke.py。此範例 SDK 呼叫尚未設定整體截止時間；試行時請以外部程序監督逾時。"},"decisions":[[{"en":"Output contract","zh-TW":"輸出契約"},{"en":"status is ok or error. data is a validated array on success (an empty array is valid), otherwise null. error is a redacted category. The envelope is ours, not the provider response.","zh-TW":"status 為 ok 或 error；成功時 data 為驗證過的陣列（可為空），失敗時為 null；error 僅保留去敏後分類。這是本站契約，不是供應商原始格式。"}],[{"en":"Reproducibility","zh-TW":"可重現性"},{"en":"Record model/version, SDK or Python version, sample identifier and environment before comparing results. Keep original media separately in authorized storage.","zh-TW":"比較結果前記錄模型／版本、SDK 或 Python 版本、樣本識別與環境；原始媒體另存於經授權儲存空間。"}],[{"en":"Failure behavior","zh-TW":"失敗處理"},{"en":"No automatic retries in these starters. Fix credentials, scopes and malformed inputs first; design bounded retries and a total deadline before production. Do not replay resource-creating requests blindly.","zh-TW":"範例不自動重試；先修正憑證、權限與輸入。正式使用前設計有限重試與整體截止時間；不可盲目重送建立資源的請求。"}],[{"en":"Acceptance gate — proposed","zh-TW":"建議驗收門檻"},{"en":"Run 20 representative authorized images plus invalid input, missing credentials and a forced timeout. Require zero silent failures. Agree a latency budget, defect recall and false-alarm threshold with the product owner before measuring; passing this smoke test alone is not production approval.","zh-TW":"以 20 張代表性授權影像，加上無效輸入、缺少憑證與強制逾時情境測試；要求零筆隱藏失敗。先與產品負責人約定延遲、缺陷召回率與誤報門檻，再量測；通過此冒煙測試不等於正式驗收。"}],[{"en":"Cost / data decision","zh-TW":"成本／資料決策"},{"en":"Both examples send image bytes to provider cloud. Confirm permitted data, retention, region and current account charges. Measure usage on a small sample before extrapolating; no cost or latency figure has been measured here.","zh-TW":"兩個範例都會將影像位元組送至供應商雲端。確認允許資料、保存期限、區域及帳號現行費用；先量測小樣本用量再推估，本頁尚無實測成本或延遲數據。"}]]},"schema_version":"1.0.0","type":"development-proposal","product_id":"industrial-api:RoboflowInference","name":"Roboflow Inference","provider":"Roboflow","url":"https://smart-tools.ai/product/RoboflowInference","localized_urls":{"en":"https://smart-tools.ai/product/RoboflowInference","zh-TW":"https://smart-tools.ai/zh-tw/product/RoboflowInference"},"category":"vision","capability_id":"cap:computer-vision","technology_hub":"https://smart-tools.ai/industrial-ai/computer-vision","interface_kind":"API / SDK","deployment":"hybrid","official_documentation":"https://docs.roboflow.com/","sources":["https://docs.roboflow.com/","https://roboflow.com/about","https://docs.roboflow.com/reference/inference/inference-sdk","https://docs.roboflow.com/reference/inference/inference-python","https://docs.roboflow.com/deployment/roboflow-cloud/serverless-api","https://docs.roboflow.com/deployment/self-hosted/inference-server","https://docs.roboflow.com/reference/inference/inference-sdk/workflows","https://docs.roboflow.com/reference/inference/inference-sdk/webrtc","https://docs.roboflow.com/reference/platform/python-sdk","https://docs.roboflow.com/reference/platform/rest-api","https://supervision.roboflow.com/latest/","https://rfdetr.roboflow.com/latest/"],"source_review":{"level":"P1","status":"official-source-reviewed","checked_at":"2026-10-08"},"provider_summary":{"en":"Deploy computer-vision models and workflows in cloud or edge environments.","zh-TW":"在雲端或邊緣裝置部署電腦視覺模型與工作流程。"},"proposed_product":{"en":"Inspection image triage","zh-TW":"巡檢影像候選缺陷清單"},"feasibility":{"tier":"prototype","rationale":{"en":"Test one documented operation with a small real sample after access is confirmed. This is a development judgment, not a delivery estimate.","zh-TW":"確認存取後，用少量真實樣本測試一個有文件的操作。此為開發判斷，不是交期估算。"},"basis":"development-judgment;not-traffic-ranked"},"proposed_inputs":{"en":"A small authorized image dataset, task definition and reference annotations.","zh-TW":"少量經授權影像、任務定義與參考標註。"},"proposed_deliverable":{"en":"A reviewable result with image references, labels or annotation state; keep original files.","zh-TW":"可審核的影像引用、標籤或標註狀態；保留原始檔案。"},"acceptance_criterion":{"en":"Compare the supported operation against a labeled sample; report errors and missing results separately.","zh-TW":"用已標註樣本比對支援的操作；錯誤與缺漏結果分開呈現。"},"dependencies":{"en":"Dataset access, image rights and a supported model or annotation project.","zh-TW":"資料集存取、影像使用權與支援的模型或標註專案。"},"integration_condition":null,"proposed_adapter_envelope":{"provider":"Roboflow","operation":"chosen-documented-operation","source_url":"https://docs.roboflow.com/","captured_at":"ISO-8601 timestamp","status":"ok|pending|unknown|failed","data":"provider response mapped by the team","evidence":"source references","errors":"explicit error details"},"adapter_envelope_status":"proposed-internal-contract;not-a-provider-response","development_status":"concept-specification","integration_test":"not-run","pricing_status":"provider-terms-review-required","estimate_status":"requires-access-and-sample-validation","traffic_status":"not-measured-in-this-assessment","measurement_events":["product_view","product_to_docs","product_to_spec","bridge_to_industrial_ai"]}