/benchmarks

TensorFlow, TensorFlow-Lite Pytorch, Torchvision, TensorRT Benchmarks

Primary LanguagePythonMIT LicenseMIT

Benchmarks

About

Benchmarks such as TensorFlow, TensorFlow-Lite, PyTorch, Torchvision, etc...

Summary

HostPC / Google Colaboratory

Description Benchmarks Framework HW
TF2 Detection Models AP, Latency TensorFlow
XLA Auto-clustering
Local PC GTX1070
TF2 Detection Models AP, Latency TensorFlow
TF-TRT
XLA Auto-clustering
Google Colab Tesla P100
TF2 Detection Models AP, Latency TensorFlow Google Colab Tesla V100
TF2 Detection Models Latency TensorFlow V100 vs P100 vs GTX1070
Torchvision detection models AP, Latency,
YouTube video
PyTorch, Torchvision Local PC GTX1070

Raspberry Pi / Coral Dev Board + EdgeTPU

Description Benchmarks Framework HW
TensorFlow 1 Detection Model Zoo AP, Latency TensorFlow Lite
XNNPACK delegate
ARM NN delegate
EdgeTPU delegate
Raspberry Pi 2, 3, 4
Coral USB Accelerator
EdgeTPU Models Latency TensorFlow Lite
EdgeTPU delegate
Coral Dev Board
EfficientDet-Lite AP, Latency TensorFlow Lit
XNNPACK delegate
ARM NN delegate
EdgeTPU delegate
Raspberry Pi 2, 3, 4
Coral USB Accelerator
YOLOX TensorFlow LiteM AP, Latency TensorFlow Lite
XNNPACK delegate
Raspberry Pi 4
TF-Lite SpaghettiNet EdgeTpu Latency vs max_detections AP, Latency TensorFlow Lite EdgeTPU delegate Raspberry Pi 4 + Coral USB Accelerator
Coral Dev Board

Jetson

Description Benchmarks Framework HW
TensorRT EfficientDet-Lite Latency,
YouTube video
TensorRT Jetson Nano
TensorRT Autoseg-EdgeTPU and
DeepLab v3+ MobilenetEdgeTPUV2
Latency,
YouTube video
TensorRT Jetson Nano
TensorRT Fast-SCNN Latency,
YouTube video
TensorRT Jetson Nano
TensorRT Ultra-Fast-Lane-Detection Latency,
YouTube video
TensorRT Jetson Nano

Reference