Skip to content

Image Classification Pipeline

The TransformersSharp.ImageClassificationPipeline class provides a high-level interface for performing image classification tasks using pre-trained models from the Hugging Face Transformers library. It simplifies the process of classifying images by handling preprocessing, model inference, and decoding.

What is an Image Classification Pipeline?

An image classification pipeline is designed to classify input images into predefined categories. It is commonly used for tasks like:

  • Object recognition
  • Scene classification
  • Image tagging

Key Features of Image Classification Pipelines:

  • Pre-trained Models: Leverages state-of-the-art models like MobileNet, ResNet, and others.
  • Top-k Results: Returns the top-k classification results with confidence scores.
  • Flexible Input: Supports both local image files and URLs.

Using the ImageClassificationPipeline Class

The ImageClassificationPipeline class in TransformersSharp provides methods to classify images. Below are examples of how to use it.

Classifying a Single Image

using TransformersSharp.Pipelines;

var pipeline = ImageClassificationPipeline.FromModel("google/mobilenet_v2_1.0_224");
var imagePath = "https://huggingface.co/datasets/Narsil/image_dummy/raw/main/parrots.png"; // Can be a local path
var result = pipeline.Classify(imagePath);
foreach (var classification in result)
{
    Console.WriteLine($"Label: {classification.Label}, Score: {classification.Score}");
}

Equivalent Python Code:

from transformers import pipeline

pipeline = pipeline("image-classification", model="google/mobilenet_v2_1.0_224")
image_path = "https://huggingface.co/datasets/Narsil/image_dummy/raw/main/parrots.png"  # Can be a local path
results = pipeline(image_path)
for item in results:
    print(f"Label: {item['label']}, Score: {item['score']}")

Classifying with Custom Parameters

using TransformersSharp.Pipelines;

var pipeline = ImageClassificationPipeline.FromModel("google/mobilenet_v2_1.0_224");
var imagePath = "https://huggingface.co/datasets/Narsil/image_dummy/raw/main/parrots.png"; // Replace with a valid image path
var result = pipeline.Classify(imagePath, topk: 3);
foreach (var classification in result)
{
    Console.WriteLine($"Label: {classification.Label}, Score: {classification.Score}");
}

Equivalent Python Code:

from transformers import pipeline

pipeline = pipeline("image-classification", model="google/mobilenet_v2_1.0_224")
image_path = "https://huggingface.co/datasets/Narsil/image_dummy/raw/main/parrots.png"  # Replace with a valid image path
results = pipeline(image_path, top_k=3)
for item in results:
    print(f"Label: {item['label']}, Score: {item['score']}")

Accessing the Tokenizer

The ImageClassificationPipeline class does not use a tokenizer, as it is designed for image inputs. Instead, it preprocesses image data directly.

Customizing Image Classification

The Classify method allows users to customize the classification process by specifying parameters like:

  • functionToApply: Specifies the function to apply to the model's output (e.g., "softmax", "sigmoid", or "none").
  • topk: Limits the number of top results to return.
  • timeout: Sets a timeout for the classification process.

Example: Customizing Parameters

using TransformersSharp.Pipelines;

var pipeline = ImageClassificationPipeline.FromModel("google/mobilenet_v2_1.0_224");
var imagePath = "https://huggingface.co/datasets/Narsil/image_dummy/raw/main/parrots.png"; // Replace with a valid image path
var result = pipeline.Classify(imagePath, functionToApply: "softmax", topk: 5);
foreach (var classification in result)
{
    Console.WriteLine($"Label: {classification.Label}, Score: {classification.Score}");
}

Equivalent Python Code:

from transformers import pipeline

pipeline = pipeline("image-classification", model="google/mobilenet_v2_1.0_224")
image_path = "https://huggingface.co/datasets/Narsil/image_dummy/raw/main/parrots.png"  # Replace with a valid image path
results = pipeline(image_path, function_to_apply="softmax", top_k=5)
for item in results:
    print(f"Label: {item['label']}, Score: {item['score']}")