Releases: Xilinx/finn-examples
FINN-examples v0.0.7 for Pynq-Z1
This release is functionally identical with FINN-examples v0.0.7 but softens the constraint on the numpy version, to allow for installation on Pynq-Z1 and Pynq-Z2 with 32-bit ARM processor.
FINN-examples v0.0.7
Updates to FINN-examples since the last release include
- New build flows and bitfiles for all models with FINN v0.10+
- DSP-packed MVAU in MobileNet-v1 on ZCU104 and VGG10-RadioML example
- Efficient implementation for per tensor thresholding, e.g. for ResNet-50 which allows the full design to be implemented on-chip without external memory
- New example for Pynq-Z1: Binarized CNV topology from the FINN paper trained on the German Traffic Sign Recognition Benchmark (GTSRB) dataset
- More stable flow for ResNet-50, fixed floorplanning and fixed FIFO depths to allow for faster rebuild times
- The opset versions of the model files were updated to allow easy integration with the verification capabilities of the builder flow
Pre-release for finn-examples
This pre-release updates the existing build flows to be compatible with FINN v0.10 (Xilinx/finn#1026) and v0.10.1 (Xilinx/finn#1128). Additionally, updated onnx files of the existing examples are published to be referenced in the final release.
FINN-examples v0.0.6
Updates to FINN-examples since the last release include
- Migration to PYNQ v3.0.1 and FINN v0.9
- Newly built bitfiles with FINN v0.9
- Migrated to new FINN (v0.9) driver: updated notebooks & QONNX dependency
- A new notebook example and build script:
- Cybersecurity MLP: classifying network packets as malicious or not with an MLP on the UNSW-NB15 dataset.
- Minor fixes in reporting and documentation
Known issues
- The build flow for ResNet-50 for the Alveo U250 has known issues and we're currently working on resolving them. However, you can still execute the associated notebook, as we provide a pre-built FPGA bitfile generated with an older Vivado (/FINN) version targeting the xilinx_u250_xdma_201830_2 platform.
- The build flow for MobileNet-v1 for the ZCU104 works, but there are currently (known) issues with runtime-writeable weights. Therefore, we have excluded this example, but you can still integrate this IP into your design and try to deploy it with your driver.
finn-examples v0.0.5
- No changes to prebuilt examples
- Update FINN to v0.8.1 for rebuilds
- Update driver and scripts to account for QONNX migration (finn-base is no longer a dependency)
- Minor fixes in reporting and documentation
finn-examples v0.0.4
Three new examples:
- Face mask wear and positioning: Low-power BNN classifier for Pynq-Z1 for correct face mask wear and positioning. Contributed by TU Munich and BMW.
- Radio signal modulation: Classify RadioML 2018.1 at 250k inferences/second on a ZCU104. See also our ITU challenge on the topic. Contributed by Felix Jentzsch.
- Keyword spotting: Classify spoken commands from the Google Speech Commands dataset at 250k inferences/second on a Pynq-Z1. Utilizes the new QONNX build flow. Contributed by Hendrik Borras.
RadioML
Data files and models for the KWS example
Data files and models for the upcoming keyword spotting (KWS) example
BinCoP face-mask detection example
New contribution from TU Munich / BMW: Binary Neural Network-based COVID-19 Face-Mask Wear and Positioning Predictor on Edge Devices
See https://github.com/NaelF/BinaryCoP/tree/master/notebook for more details
Note: currently PYNQ-Z1 only, bitfile only
finn-examples v0.0.3b
- ResNet-50: first release on U250