EIDORA/ResNet50_IN1k

ResNet50 Image Light

ResNet-50 is a residual convolutional network. EIDORA uses its pretrained ImageNet features as normalized 2048-dimensional image embeddings for collection analysis.

Training Data and Checkpoint Provenance

• Architecture: ResNet-50 residual CNN with global average pooling.
• Original checkpoint: torchvision/resnet50-imagenet1k-v2.
• Weight variant: IMAGENET1K_V2 (TorchVision download).
• Training data: ImageNet-1K
• Training method: Supervised image classification.
• Source: ResNet-50 (original author attribution retained).

Model Overview

• Model: ResNet50 Image Light
• EIDORA identifier: eidora/image-resnet50-imagenet-light
• Modality: Image
• Embedding dimensions: 2048
• 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:

• Explore image similarity and build visual clusters.
• Compare residual CNN embeddings against AlexNet, VGG16, DINO or CLIP.
• Create a reproducible ImageNet-based baseline for image collections.

Input and Preprocessing

EIDORA prepares the image before inference:

1. Convert to RGB.
2. Resize the shorter side to 232 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', 2048]. Each vector has 2048 dimensions, is L2-normalized, and is intended for cosine similarity comparisons.

EIDORA removes the final classification prediction and uses the global average pooled 2048-dimensional feature vector, with L2 normalization.

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 original task is ImageNet classification, not historical or cultural interpretation.
• Fine-grained meaning and identity cannot be inferred from feature proximity alone.
• The EIDORA resize setting is part of the embedding definition.

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

Research Context and References

Main paper

Deep Residual Learning for Image Recognition

Paper: https://arxiv.org/abs/1512.03385

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)

@inproceedings{he2016deep,
  title={Deep Residual Learning for Image Recognition},
  author={He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian},
  booktitle={CVPR},
  year={2016}
}

License and Attribution

The TorchVision software library uses BSD-3-Clause terms. The linked upstream model information does not establish that the pretrained ImageNet weights carry that same license. No separate weight-license identifier is asserted here. EIDORA provides a conversion of the original TorchVision weights, not a newly trained ResNet-50.

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/ResNet50_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.