EIDORA/VGG16_IN1k

VGG16 Image Medium

VGG16 is a deep convolutional image classifier. This EIDORA variant exposes its 4096-dimensional classifier features as normalized image embeddings.

Training Data and Checkpoint Provenance

• Architecture: VGG16 convolutional network with fully connected classifier.
• Original checkpoint: torchvision/vgg16-imagenet1k-v1.
• Weight variant: IMAGENET1K_V1 (TorchVision download).
• Training data: ImageNet-1K
• Training method: Supervised image classification.
• Source: VGG-16 (original author attribution retained).

Model Overview

• Model: VGG16 Image Medium
• EIDORA identifier: eidora/image-vgg16-imagenet-medium
• Modality: Image
• Embedding dimensions: 4096
• Embedding normalization: L2-normalized
• Similarity measure: Cosine similarity
• Package schema: Version 3
• Model version: 0.1.0

EIDORA Use and Supported Modalities

Supported modality: Images only.

Researchers can use this package to:

• Compare classical VGG features with newer visual embeddings.
• Explore visual clustering, similarity and retrieval.
• Use a historical CNN baseline when studying image representations.

Input and Preprocessing

EIDORA prepares the image before inference:

1. Convert to RGB.
2. Resize the shorter side to 256 pixels, then take a center crop of 224 × 224.
3. Use bilinear interpolation. Use truncate dimension rounding.
4. Rescale by 0.00392156862745098 and normalize each RGB channel.
5. Arrange the tensor in NCHW layout and pass float32 pixel_values with shape [batch, 3, 224, 224].

Mean: [0.485, 0.456, 0.406]
Standard deviation: [0.229, 0.224, 0.225]

Normalization is performed by EIDORA and should not be applied a second time. The defaults in config.yaml determine the documented embedding behavior; alternate settings may change the vectors.

Output Embeddings

The ONNX output is float32 with shape ['batch', 4096]. Each vector has 4096 dimensions, is L2-normalized, and is intended for cosine similarity comparisons.

EIDORA extracts the second fully connected layer after ReLU and applies L2 normalization, rather than returning ImageNet class scores.

Runtime and Compatibility

The package uses model.onnx and schema-v3 runtime declarations for image preparation and embedding output.

It declares CPUExecutionProvider. Actual execution availability depends on the installed ONNX Runtime and EIDORA Desktop environment. config.yaml is authoritative for graph inputs, preprocessing, provider and embedding dimensions.

Limitations

• The checkpoint was trained for ImageNet classification, not specialized research collections.
• The ONNX weights and memory footprint are relatively large.
• Cosine similarity of features does not establish image provenance or historical meaning.

Results should be evaluated against the intended research data and questions.

Research Context and References

Main paper

Very Deep Convolutional Networks for Large-Scale Image Recognition

Paper: https://arxiv.org/abs/1409.1556

The publication describes the upstream learning method and original model. The embeddings returned by EIDORA additionally depend on the specific preprocessing and output adaptation documented here.

Primary citation (BibTeX)

@article{simonyan2014very,
  title={Very Deep Convolutional Networks for Large-Scale Image Recognition},
  author={Simonyan, Karen and Zisserman, Andrew},
  journal={arXiv preprint arXiv:1409.1556},
  year={2014}
}

License and Attribution

TorchVision source code uses BSD-3-Clause terms; pretrained ImageNet weights may be governed by additional terms. The linked source documentation does not establish a separate license for the checkpoint weights, so no such license is claimed. The EIDORA model is an ONNX conversion of those weights.

No separate upstream checkpoint LICENSE file was supplied in the captured EIDORA repository; that absence should not be interpreted as permission or prohibition.

Package Information

• Model repository: EIDORA/VGG16_IN1k
• Package version: 0.1.0
• Package schema: 3
• EIDORA configuration: config.yaml
• ONNX graph(s): model.onnx
• Supported modality: Image
• License identifier: other
• Declared provider: CPUExecutionProvider

The configuration defines the authoritative model-facing runtime contract. Other packaged assets, where present, do not introduce additional inference modalities.