@camstack/addon-scene-intelligence 0.1.3 → 0.1.5

package/dist/index.d.ts

@@ -1,87 +0,0 @@
-import { ICamstackAddon, ISceneIntelligence, AddonManifest, AddonContext, CapabilityProviderMap, CropInput, ClassifierOutput, EmbeddingMetadata, EmbeddingFilter, VectorSearchResult, SceneStateResult, IScopedLogger, IEmbeddingsBackend, ModelCatalogEntry } from '@camstack/types';
-
-declare class SceneIntelligenceAddon implements ICamstackAddon, ISceneIntelligence {
-    readonly manifest: AddonManifest;
-    private logger;
-    private imageEncoder;
-    private textEncoder;
-    private sceneStateMachine;
-    private referenceStore;
-    private searchService;
-    private ctx;
-    initialize(context: AddonContext): Promise<void>;
-    shutdown(): Promise<void>;
-    getCapabilityProvider<K extends keyof CapabilityProviderMap>(name: K): CapabilityProviderMap[K] | null;
-    classify(input: CropInput): Promise<ClassifierOutput>;
-    embed(deviceId: string, crop: Buffer, metadata: EmbeddingMetadata): Promise<string>;
-    search(query: string, topK: number, filter?: EmbeddingFilter): Promise<VectorSearchResult[]>;
-    searchByImage(image: Buffer, topK: number, filter?: EmbeddingFilter): Promise<VectorSearchResult[]>;
-    evaluateSceneState(deviceId: string, crop: Buffer): Promise<SceneStateResult | null>;
-    private ensureTextEncoder;
-}
-
-declare class ClipImageEncoder {
-    private session;
-    private readonly modelId;
-    private readonly inputSize;
-    private readonly logger;
-    constructor(modelId: string, logger: IScopedLogger);
-    load(modelPath: string): Promise<void>;
-    /**
-     * Encode a raw RGB buffer into a CLIP embedding.
-     * Caller must provide RGB buffer (use sharp to decode JPEG first).
-     */
-    encode(rgb: Buffer, width: number, height: number): Promise<Float32Array>;
-    dispose(): Promise<void>;
-}
-
-declare class ClipTextEncoder {
-    private session;
-    private readonly logger;
-    constructor(logger: IScopedLogger);
-    load(modelPath: string): Promise<void>;
-    encode(text: string): Promise<Float32Array>;
-    dispose(): Promise<void>;
-}
-
-declare class SearchService {
-    private readonly backend;
-    private index;
-    constructor(backend: IEmbeddingsBackend);
-    initialize(): Promise<void>;
-    storeEmbedding(id: string, embedding: Float32Array, metadata: EmbeddingMetadata): Promise<void>;
-    searchByVector(query: Float32Array, topK: number, filter?: EmbeddingFilter): Promise<readonly VectorSearchResult[]>;
-    count(): Promise<number>;
-    shutdown(): Promise<void>;
-}
-
-interface SceneStateDefinition {
-    readonly id: string;
-    readonly name: string;
-    readonly referenceEmbedding: Float32Array;
-    readonly threshold: number;
-}
-declare class SceneStateMachine {
-    private readonly debounceFrames;
-    private readonly cameraStates;
-    constructor(debounceFrames?: number);
-    evaluate(deviceId: string, embedding: Float32Array, referenceStates: readonly SceneStateDefinition[]): SceneStateResult | null;
-}
-
-declare const CLIP_MODELS: readonly ModelCatalogEntry[];
-declare const DEFAULT_CLIP_MODEL = "mobileclip-s0";
-declare const CLIP_EMBEDDING_DIM = 512;
-
-/**
- * Preprocess raw RGB buffer for CLIP inference.
- * Resizes (nearest-neighbor for speed), normalizes with CLIP mean/std, outputs NCHW Float32Array.
- * For production use, the caller should use sharp to resize the JPEG to targetW×targetH
- * before calling this with the raw RGB. This function handles normalization + layout.
- */
-declare function preprocessForClip(rgb: Buffer, srcWidth: number, srcHeight: number, targetWidth: number, targetHeight: number): Float32Array;
-/**
- * L2-normalize a vector in-place and return it.
- */
-declare function l2Normalize(vec: Float32Array): Float32Array;
-
-export { CLIP_EMBEDDING_DIM, CLIP_MODELS, ClipImageEncoder, ClipTextEncoder, DEFAULT_CLIP_MODEL, SceneIntelligenceAddon, SceneStateMachine, SearchService, SceneIntelligenceAddon as default, l2Normalize, preprocessForClip };