Pasar al contenido principal

Resultados relacionados

Resultados totales (Incluyendo duplicados): 2
Encontrada(s) 1 página(s)

UCrea Repositorio Abierto de la Universidad de Cantabria
Publicación Contribución de congreso (ConferenceOutput)

CNN-LSTM implementation methodology on SoC FPGA for human action recognition based on video

  • Suárez Plata, Daniel Nicolás
  • Fernández Solórzano, Víctor Manuel|||0000-0003-0614-151X
  • Posadas Cobo, Héctor
The growing use of AI -driven video applications like surveillance or healthcare monitoring underscores the need for embedded solutions capable of accurately categorizing human actions in real-time videos. A methodology is proposed for implementing a customized CNN-LSTM architecture on AMD-Xilinx SoC FPGA devices for human action categorization from video data. In this approach, CNN operations are accelerated by the Vitis-AI DPU within the FPG A, offering flexibility to support a range of CNN architectures without requiring individual hardware description language development. This adaptability is crucial given the varying performance of CNN models across datasets. LSTM operations are executed on the SoC processors, overcoming limitations in the support provided by DPU IP cores for such networks, while maintaining flexibility to assess different configurations. Additionally, a pipeline strategy is proposed to enable parallel execution of both CNN and LSTM components, optimizing resource utilization and minimizing idle times. To demonstrate the validity of the proposed implementation methodology, experiments were conducted on the ZCUI02 de-velopment board, equipped with a Zynq Ultrascale+ MP-SoC, and involved the use of the VGG 16 CNN model along with the exploration of different LSTM configurations. The results demonstrate remarkable computational performance, achieving frame rates of up to 44.34 FPS for videos recorded at a resolution of 320×240 pixels, surpassing real-time requirements. Aditionally, the proposed implementation maintains high accuracy levels, exemplified by the single bidirectional LSTM layer achieving a competitive accuracy of 73.33% based on the UCF10l dataset., This work has been supported by Project PID2020-116417RB-C43, funded by Spanish MCIN/AEI/10.13039/501100011033 and by Project No 101007273 ECSEL DAIS, funded by EU H2020 and by Spanish pci2021-121988.
UCrea Repositorio Abierto de la Universidad de Cantabria
Publicación Contribución de congreso (ConferenceOutput)

UML-based design flow for systems with neural networks

  • Suárez Plata, Daniel Nicolás
  • Posadas Cobo, Héctor
  • Fernández Solórzano, Víctor Manuel|||0000-0003-0614-151X
Artificial intelligence has demonstrated its ability to solve lots of critical tasks, but at the cost of high computational requirements. Different hardware has been proposed to provide this computational power, each one with its benefits and drawbacks. However, the exploration of the different alternatives in an easy an integrated way is still a complex task. To solve so, this paper proposes a UML-based design flow where neural networks are initially specified and then automatically generated and trained using TensorFlow. The approach also enables automatic mapping of models to CPU, GPU and FPGAs, using Xilinx’s Deep Learning Processor Units (DPUs). The framework also generates the communication codes required to connect the other system components with the implementation selected. This approach addresses design-space exploration challenges, system architecture definition, and improves implementation and training processes by saving time and effort., This work has been supported by Project PID2020-116417RB-C43, funded by Spanish MCIN/AEI/10.13039/501100011033 and by the KDT JU agreement No 101007273 ECSEL DAIS, funded by EU H2020 and by Spanish pci2021- 121988