Xinet_picasso_muse

Use case : Neural style transfer

Model description

Xinet_picasso_muse is a lightweight Neural Style Transfer approach based on XiNets, neural networks especially developed for microcontrollers and embedded applications. It has been trained using the COCO dataset for content images and the painting La Muse of Pablo Picasso for style image. This model achieves an extremely lightweight transfer style mechanism and high-quality stylized outputs, significantly reducing computational complexity.

Xinet_picasso_muse is implemented initially in Pytorch and is quantized in int8 format using tensorflow lite converter. To reach a better performances, the mirror padding ops have been replaced with zero padding ops.

Network information

Recommended platform

---

Performances

Metrics

Measures are done with default STEdgeAI Core configuration with enabled input / output allocated option.

Reference NPU memory footprint based on COCO dataset

Reference NPU inference time based on COCO Person dataset

Retraining and Integration in a Simple Example

Retraining and deployment services are currently not provided for this model. They should be supported in the future releases.

References

[1] "Painting the Starry Night using XiNets" Alberto Ancilotto, Elisabetta Farella - 2024 IEEE International Conference on Pervasive Computing Link