EfficientNet is a state-of-the-art image classification model that has been trained on the ImageNet-1k dataset using innovative techniques like Noisy Student semi-supervised learning. In this blog, we will walk you through how to use the EfficientNet model for image classification tasks, feature map extraction, and generating image embeddings.

Model Overview

• Model Type: Image classification feature backbone
• Model Stats:
  • Parameters (M): 5.3
  • GMACs: 0.4
  • Activations (M): 6.7
  • Image Size: 224 x 224
• Dataset: ImageNet-1k
• Papers:
  • EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
  • Self-training with Noisy Student Improves ImageNet Classification
• Original Codebase: GitHub Repository

Model Usage

Below are the steps to use the EfficientNet model for various tasks.

1. Image Classification

Follow these steps to classify an image:

python
from urllib.request import urlopen
from PIL import Image
import timm

# Load image from URL
img = Image.open(urlopen("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png"))

# Initialize the model
model = timm.create_model('tf_efficientnet_b0.ns_jft_in1k', pretrained=True)
model = model.eval()

# Get model-specific transforms (normalization, resize)
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)

# Perform classification
output = model(transforms(img).unsqueeze(0))
top5_probabilities, top5_class_indices = torch.topk(output.softmax(dim=1) * 100, k=5)

In the code above, think of the EfficientNet model as a highly trained librarian who knows exactly how to categorize every book (image) that comes through the door. The librarian takes a book, applies the appropriate dust jacket (transforms), and then provides you with the top 5 categories it thinks the book belongs to based on its extensive training.

2. Feature Map Extraction

To extract feature maps from an image, use the following code:

python
# Load image
img = Image.open(urlopen("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png"))

# Initialize the model with features_only
model = timm.create_model('tf_efficientnet_b0.ns_jft_in1k', pretrained=True, features_only=True)
model = model.eval()

# Get model-specific transforms
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)

# Perform feature extraction
output = model(transforms(img).unsqueeze(0))

for o in output:
    # Shape of each feature map
    print(o.shape)

In this analogy, think of feature maps as different layers of filters that a photographer applies to a photograph before presenting it to an audience. Each filter highlights various aspects of the image, helping to identify important features regardless of the context.

3. Image Embeddings

To generate embeddings for an image, utilize the following code:

python
# Load image
img = Image.open(urlopen("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png"))

# Initialize the model without the classifier
model = timm.create_model('tf_efficientnet_b0.ns_jft_in1k', pretrained=True, num_classes=0)
model = model.eval()

# Perform embedding generation
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)
output = model(transforms(img).unsqueeze(0))

# Output contains unpooled features
output = model.forward_features(transforms(img).unsqueeze(0))
output = model.forward_head(output, pre_logits=True)

Here, image embeddings can be regarded as a unique fingerprint of the photograph that encapsulates its most critical features, allowing for precise identification and categorization.

Troubleshooting

If you encounter issues while implementing the model, here are some troubleshooting tips:

• Ensure you have installed all the required libraries, primarily timm and PIL.
• Check the image URL for correctness. If the image is unavailable, try using a different source.
• Ensure that your environment is compatible with the PyTorch version required for the model.
• Monitor your RAM and GPU memory usage; large models may require more resources than available.

For more insights, updates, or to collaborate on AI development projects, stay connected with fxis.ai.

Final Words

At fxis.ai, we believe that such advancements are crucial for the future of AI, as they enable more comprehensive and effective solutions. Our team is continually exploring new methodologies to push the envelope in artificial intelligence, ensuring that our clients benefit from the latest technological innovations.