RegNet

Use case : Image classification

Model description

RegNet introduces a design space paradigm for neural networks. Rather than designing individual architectures, RegNet defines a design space of possible networks characterized by a few parameters, enabling systematic exploration of network designs.

The architecture uses quantized linear parameterization where networks are defined by simple equations, with systematic variation of width and depth patterns. RegNet employs group convolutions for efficiency, following a design space exploration methodology for finding optimal configurations.

RegNet is well-suited for research on neural network design principles, applications requiring systematic architecture selection, and scalable deployments with consistent design principles.

(source: https://arxiv.org/abs/2003.13678)

The model is quantized to int8 using ONNX Runtime and exported for efficient deployment.

Network information

Network inputs / outputs

For an image resolution of NxM and P classes

Recommended platforms

Performances

Metrics

• Measures are done with default STEdgeAI Core configuration with enabled input / output allocated option.
• All the models are trained from scratch on Imagenet dataset

Reference NPU memory footprint on Imagenet dataset (see Accuracy for details on dataset)

Reference NPU inference time on Imagenet dataset (see Accuracy for details on dataset)

Accuracy with Imagenet dataset

Dataset details: link Number of classes: 1000. To perform the quantization, we calibrated the activations with a random subset of the training set. For the sake of simplicity, the accuracy reported here was estimated on the 50000 labelled images of the validation set.

Retraining and Integration in a simple example:

Please refer to the stm32ai-modelzoo-services GitHub here

References

[1] - Dataset: Imagenet (ILSVRC 2012) — https://www.image-net.org/

[2] - Model: RegNet — https://github.com/facebookresearch/pycls