HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs

20 Jul 2020Hai Victor HabiRoy H. JenningsArnon Netzer

Recent work in network quantization produced state-of-the-art results using mixed precision quantization. An imperative requirement for many efficient edge device hardware implementations is that their quantizers are uniform and with power-of-two thresholds... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Quantization ImageNet EfficientNet-B0-W8A8 Accuracy (%) 76.4 # 1
Quantization ImageNet EfficientNet-B0-W4A4 Accuracy (%) 76 # 2
Quantization ImageNet ResNet50-W3A4 Accuracy (%) 75.45 # 3
Quantization ImageNet MobileNetV2 Accuracy (%) 70.9 # 6

Methods used in the Paper