ResNet

Use case : Image classification

Model description

Residual Network (ResNet) introduced skip connections that enable training of very deep networks. It revolutionized deep learning by solving the degradation problem in deep networks.

ResNet features skip connections that add input to output, enabling gradient flow, with the network learning residual functions with reference to layer inputs. Batch normalization is applied after every convolution for stable training, and a bottleneck design uses 1x1-3x3-1x1 convolution patterns for efficiency.

ResNet serves as the baseline for computer vision tasks, a transfer learning source model, and is widely used for research and benchmarking.

(source: https://arxiv.org/abs/1512.03385)

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

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: ResNet — https://github.com/KaimingHe/deep-residual-networks