[version] Update to v1.4.2

Author: xenova

dist/transformers.js

@@ -1,8 +1,81 @@
-/**
- *
- * @param {string} url
- * @returns {Promise<any>}
- */
-export function loadImage(url: string): Promise<any>;
-export const Jimp: any;
-export const ImageType: any;
+export class CustomImage {
+    /**
+     * Helper method for reading an image from a variety of input types.
+     * @param {CustomImage|string|URL} input
+     * @returns The image object.
+     */
+    static read(input: CustomImage | string | URL): Promise<CustomImage>;
+    /**
+     * Read an image from a URL or file path.
+     * @param {string|URL} url - The URL or file path to read the image from.
+     * @returns {Promise<CustomImage>} - The image object.
+     */
+    static fromURL(url: string | URL): Promise<CustomImage>;
+    /**
+     * Helper method to create a new canvas, draw an image/canvas to it, then return the pixel data.     * @param {ImageClass|CanvasClass} img - The image/canvas to draw to the canvas.
+     * @param {number} [width=null] - Width of the canvas. If null, the width of the image is used.
+     * @param {number} [height=null] - Height of the canvas. If null, the height of the image is used.
+     * @returns {CustomImage} - The image object.
+     */
+    static createCanvasAndDraw(img: any, width?: number, height?: number): CustomImage;
+    /**
+     * Create a new CustomImage object.
+     * @param {Uint8ClampedArray} data - The pixel data.
+     * @param {number} width - The width of the image.
+     * @param {number} height - The height of the image.
+     * @param {number} channels - The number of channels.
+     */
+    constructor(data: Uint8ClampedArray, width: number, height: number, channels: number);
+    /**
+     * Convert the image to grayscale format.
+     * @returns {CustomImage} - `this` to support chaining.
+     */
+    grayscale(): CustomImage;
+    /**
+     * Convert the image to RGB format.
+     * @returns {CustomImage} - `this` to support chaining.
+     */
+    rgb(): CustomImage;
+    /**
+     * Convert the image to RGBA format.
+     * @returns {CustomImage} - `this` to support chaining.
+     */
+    rgba(): CustomImage;
+    /**
+     * Resize the image to the given dimensions. This method uses the canvas API to perform the resizing.
+     * @param {number} width - The width of the new image.
+     * @param {number} height - The height of the new image.
+     * @returns {CustomImage} - `this` to support chaining.
+     */
+    resize(width: number, height: number): CustomImage;
+    toCanvas(): any;
+    /**
+     * Helper method to update the image data.
+     * @param {Uint8ClampedArray} data - The new image data.
+     * @param {number} width - The new width of the image.
+     * @param {number} height - The new height of the image.
+     * @param {number} channels - The new number of channels of the image.
+     */
+    _update(data: Uint8ClampedArray, width: number, height: number, channels?: number): CustomImage;
+    data: Uint8ClampedArray;
+    width: number;
+    height: number;
+    channels: number;
+    /**
+     * Clone the image
+     * @returns {CustomImage} - The cloned image
+     */
+    clone(): CustomImage;
+    /**
+     * Helper method for converting image to have a certain number of channels
+     * @param {number} numChannels - The number of channels. Must be 1, 3, or 4.
+     * @returns {CustomImage} - `this` to support chaining.
+     */
+    convert(numChannels: number): CustomImage;
+    /**
+     * Save the image to the given path. This method is only available in environments with access to the FileSystem.
+     * @param {string|Buffer|URL} path - The path to save the image to.
+     * @param {string} [mime='image/png'] - The mime type of the image.
+     */
+    save(path: string | Buffer | URL, mime?: string): void;
+}

dist/transformers.js.map

@@ -0,0 +1,22 @@
+export type TypedArray = Int8Array | Uint8Array | Uint8ClampedArray | Int16Array | Uint16Array | Int32Array | Uint32Array | Float32Array | Float64Array;
+export type BigTypedArray = BigInt64Array | BigUint64Array;
+export type AnyTypedArray = TypedArray | BigTypedArray;
+/**
+ * @typedef {Int8Array | Uint8Array | Uint8ClampedArray | Int16Array | Uint16Array | Int32Array | Uint32Array | Float32Array | Float64Array} TypedArray
+ * @typedef {BigInt64Array | BigUint64Array} BigTypedArray
+ * @typedef {TypedArray | BigTypedArray} AnyTypedArray
+ */
+/**
+ * @param {TypedArray} input
+ */
+export function interpolate(input: TypedArray, [in_channels, in_height, in_width]: [any, any, any], [out_height, out_width]: [any, any], mode?: string, align_corners?: boolean): any;
+/**
+ * Helper method to transpose a AnyTypedArray directly
+ * @param {T} array
+ * @template {AnyTypedArray} T
+ * @param {number[]} dims
+ * @param {number[]} axes
+ * @returns {[T, number[]]} The transposed array and the new shape.
+ */
+declare function transpose_data<T extends AnyTypedArray>(array: T, dims: number[], axes: number[]): [T, number[]];
+export { transpose_data as transpose };

dist/transformers.min.js

@@ -175,7 +175,7 @@ export class AutoModelForImageClassification {
      * Loads a pre-trained image classification model from a given directory path.
      * @param {string} modelPath - The path to the directory containing the pre-trained model.
      * @param {function} [progressCallback=null] - A callback function to monitor the loading progress.
-     * @returns {Promise<PreTrainedModel>} A Promise that resolves with an instance of the ViTForImageClassification class.
+     * @returns {Promise<PreTrainedModel>} A Promise that resolves with the model.
      * @throws {Error} If the specified model type is not supported.
      */
     static from_pretrained(modelPath: string, progressCallback?: Function): Promise<PreTrainedModel>;
@@ -188,7 +188,23 @@ export class AutoModelForObjectDetection {
      * Loads a pre-trained image classification model from a given directory path.
      * @param {string} modelPath - The path to the directory containing the pre-trained model.
      * @param {function} [progressCallback=null] - A callback function to monitor the loading progress.
-     * @returns {Promise<PreTrainedModel>} A Promise that resolves with an instance of the ViTForImageClassification class.
+     * @returns {Promise<PreTrainedModel>} A Promise that resolves with the model.
+     * @throws {Error} If the specified model type is not supported.
+     */
+    static from_pretrained(modelPath: string, progressCallback?: Function): Promise<PreTrainedModel>;
+}
+/**
+ * AutoModelForImageSegmentation is a class for loading pre-trained image classification models from ONNX format.
+ */
+export class AutoModelForImageSegmentation {
+    static MODEL_CLASS_MAPPING: {
+        detr: typeof DetrForSegmentation;
+    };
+    /**
+     * Loads a pre-trained image classification model from a given directory path.
+     * @param {string} modelPath - The path to the directory containing the pre-trained model.
+     * @param {function} [progressCallback=null] - A callback function to monitor the loading progress.
+     * @returns {Promise<PreTrainedModel>} A Promise that resolves with the model.
      * @throws {Error} If the specified model type is not supported.
      */
     static from_pretrained(modelPath: string, progressCallback?: Function): Promise<PreTrainedModel>;
@@ -1039,6 +1055,14 @@ declare class DetrForObjectDetection extends DetrPreTrainedModel {
      */
     _call(model_inputs: any): Promise<DetrObjectDetectionOutput>;
 }
+declare class DetrForSegmentation extends DetrPreTrainedModel {
+    /**
+     * Runs the model with the provided inputs
+     * @param {Object} model_inputs - Model inputs
+     * @returns {Promise<DetrSegmentationOutput>} - Object containing segmentation outputs
+     */
+    _call(model_inputs: any): Promise<DetrSegmentationOutput>;
+}
 declare class BertPreTrainedModel extends PreTrainedModel {
 }
 declare class AlbertPreTrainedModel extends PreTrainedModel {
@@ -1127,6 +1151,17 @@ declare class DetrObjectDetectionOutput extends ModelOutput {
     logits: any;
     pred_boxes: any;
 }
+declare class DetrSegmentationOutput extends ModelOutput {
+    /**
+     * @param {Tensor} logits - The output logits of the model.
+     * @param {Tensor} pred_boxes - Predicted boxes.
+     * @param {Tensor} pred_masks - Predicted masks.
+     */
+    constructor(logits: Tensor, pred_boxes: Tensor, pred_masks: Tensor);
+    logits: Tensor;
+    pred_boxes: Tensor;
+    pred_masks: Tensor;
+}
 declare class ModelOutput {
 }
 export {};

dist/transformers.min.js.map

@@ -17,7 +17,7 @@ export class AutoProcessor {
  * Represents a Processor that extracts features from an input.
  * @extends Callable
  */
-declare class Processor extends Callable {
+export class Processor extends Callable {
     /**
      * Creates a new Processor with the given feature extractor.
      * @param {function} feature_extractor - The function used to extract features from the input.
@@ -33,4 +33,3 @@ declare class Processor extends Callable {
     _call(input: any): Promise<any>;
 }
 import { Callable } from "./utils.js";
-export {};

dist/types/image_utils.d.ts

@@ -2,16 +2,22 @@
  * This creates a nested array of a given type and depth (see examples).
  */
 export type NestArray<T, Depth extends number, Acc extends never[] = []> = Acc['length'] extends Depth ? T : NestArray<T[], Depth, [...Acc, never]>;
+export type AnyTypedArray = import('./math_utils.js').AnyTypedArray;
 declare const Tensor_base: any;
 export class Tensor extends Tensor_base {
     [x: string]: any;
-    constructor(...args: any[]);
+    /**
+     * Create a new Tensor or copy an existing Tensor.
+     * @param  {[string, Array|AnyTypedArray, number[]]|[ONNXTensor]} args
+     */
+    constructor(...args: [string, any[] | AnyTypedArray, number[]] | [any]);
     /**
      *
      * @param {number} index
-     * @returns
+     * @returns {Tensor}
+     * @todo Set type based on dims
      */
-    get(index: number): any;
+    get(index: number): Tensor;
     /**
      * @param {any} item
      * @returns {number}
@@ -25,6 +31,23 @@ export class Tensor extends Tensor_base {
      */
     _subarray(index: number, iterSize: number, iterDims: any): Tensor;
     tolist(): any;
+    /**
+     * Return a new Tensor the sigmoid function applied to each element.
+     * @returns {Tensor} - The tensor with the sigmoid function applied.
+     */
+    sigmoid(): Tensor;
+    /**
+     * Applies the sigmoid function to the tensor in place.
+     * @returns {Tensor} - Returns `this`.
+     */
+    sigmoid_(): Tensor;
+    clone(): Tensor;
+    /**
+     * Return a transposed version of this Tensor, according to the provided dimensions.
+     * @param  {...number} dims - Dimensions to transpose.
+     * @returns {Tensor} - The transposed tensor.
+     */
+    transpose(...dims: number[]): Tensor;
     /**
      * Returns an iterator object for iterating over the tensor data in row-major order.
      * If the tensor has more than one dimension, the iterator will yield subarrays.
@@ -46,4 +69,14 @@ export function transpose(tensor: any, axes: any[]): Tensor;
  * @returns {Tensor} - The concatenated tensor.
  */
 export function cat(tensors: any): Tensor;
-export {};
+/**
+ * Interpolates an Tensor to the given size.
+ * @param {Tensor} input - The input tensor to interpolate. Data must be channel-first (i.e., [c, h, w])
+ * @param {number[]} size - The output size of the image
+ * @param {string} mode - The interpolation mode
+ * @param {boolean} align_corners - Whether to align corners.
+ * @returns {Tensor} - The interpolated tensor.
+ */
+export function interpolate(input: Tensor, [out_height, out_width]: number[], mode?: string, align_corners?: boolean): Tensor;
+import { transpose as transpose_data } from "./math_utils.js";
+export { transpose_data };

dist/types/math_utils.d.ts

@@ -15,5 +15,6 @@ import { AutoModelForImageClassification } from "./models.js";
 import { AutoModelForObjectDetection } from "./models.js";
 import { AutoProcessor } from "./processors.js";
 import { pipeline } from "./pipelines.js";
+import { Tensor } from "./tensor_utils.js";
 import { env } from "./env.js";
-export { AutoTokenizer, BertTokenizer, DistilBertTokenizer, T5Tokenizer, GPT2Tokenizer, AutoModel, AutoModelForSeq2SeqLM, AutoModelForSequenceClassification, AutoModelForTokenClassification, AutoModelForCausalLM, AutoModelForMaskedLM, AutoModelForQuestionAnswering, AutoModelForVision2Seq, AutoModelForImageClassification, AutoModelForObjectDetection, AutoProcessor, pipeline, env };
+export { AutoTokenizer, BertTokenizer, DistilBertTokenizer, T5Tokenizer, GPT2Tokenizer, AutoModel, AutoModelForSeq2SeqLM, AutoModelForSequenceClassification, AutoModelForTokenClassification, AutoModelForCausalLM, AutoModelForMaskedLM, AutoModelForQuestionAnswering, AutoModelForVision2Seq, AutoModelForImageClassification, AutoModelForObjectDetection, AutoProcessor, pipeline, Tensor, env };