DiAD: A Diffusion-based Framework for Multi-class Anomaly Detection

1銆丅ackground

褰撳墠涓绘祦鐨勪笁绉嶅紓甯告娴嬫柟娉曪細

• 鍩轰簬鍚堟垚鐨勬柟娉曪細鍦ㄦ甯稿浘鍍忎笂鍚堟垚寮傚父锛岄€氳繃璁粌妯″瀷璇嗗埆杩欎簺鍚堟垚鐨勫紓甯告潵鎻愰珮妫€娴嬪拰瀹氫綅鑳藉姏銆?/li>
• 鍩轰簬宓屽叆鐨勬柟娉曪細灏嗗浘鍍忎俊鎭紪鐮佸埌鐗瑰緛绌洪棿涓紝浣跨敤棰勮缁冪殑缃戠粶鎻愬彇鐗瑰緛锛屽苟璁＄畻鐗瑰緛鐨勭浉浼兼€т綔涓哄紓甯稿垎鏁般€?/li>
• 鍩轰簬閲嶅缓鐨勬柟娉曪細璁粌妯″瀷鍦ㄦ甯告暟鎹泦涓婂涔犳ā寮忓拰鐗瑰緛锛岀劧鍚庡湪娴嬭瘯闃舵閲嶅缓寮傚父鍥惧儚锛岄€氳繃姣旇緝閲嶅缓鍥惧儚鍜岃緭鍏ュ浘鍍忔潵瀹氫綅寮傚父銆?/li>

鏈€杩戯紝鎵╂暎妯″瀷灞曠ず浜嗗畠浠己澶х殑鍥惧儚鐢熸垚鑳藉姏銆傜劧鑰岋紝鐩存帴浣跨敤褰撳墠涓绘祦鐨勬墿鏁ｆā鍨嬫棤娉曟湁鏁堣В鍐?strong>澶氱被鍒紓甯告娴?/strong>闂銆?/p>