ML Research Wiki / Benchmarks / Quantization / ImageNet

ImageNet

Quantization Benchmark

Performance Over Time

📊 Showing 27 results | 📏 Metric: Top-1 Accuracy (%)

Top Performing Models

Rank	Model	Paper	Top-1 Accuracy (%)	Date	Code
1	FQ-ViT (ViT-L)	FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer	85.03	2021-11-27	📦 megvii-research/FQ-ViT
2	FQ-ViT (ViT-B)	FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer	83.31	2021-11-27	📦 megvii-research/FQ-ViT
3	FQ-ViT (Swin-B)	FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer	82.97	2021-11-27	📦 megvii-research/FQ-ViT
4	FQ-ViT (Swin-S)	FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer	82.71	2021-11-27	📦 megvii-research/FQ-ViT
5	FQ-ViT (DeiT-B)	FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer	81.20	2021-11-27	📦 megvii-research/FQ-ViT
6	FQ-ViT (Swin-T)	FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer	80.51	2021-11-27	📦 megvii-research/FQ-ViT
7	FQ-ViT (DeiT-S)	FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer	79.17	2021-11-27	📦 megvii-research/FQ-ViT
8	Xception W8A8	HPTQ: Hardware-Friendly Post Training Quantization	78.97	2021-09-19	📦 sony/model_optimization
9	ADLIK-MO-ResNet50-W4A4	Learned Step Size Quantization	77.88	2019-02-21	📦 zhutmost/lsq-net 📦 hustzxd/LSQuantization 📦 ZouJiu1/LSQplus
10	ADLIK-MO-ResNet50-W3A4	Learned Step Size Quantization	77.34	2019-02-21	📦 zhutmost/lsq-net 📦 hustzxd/LSQuantization 📦 ZouJiu1/LSQplus

All Papers (27)

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

2021

FQ-ViT (ViT-L)

megvii-research/FQ-ViT

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

2021

FQ-ViT (ViT-B)

megvii-research/FQ-ViT

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

2021

FQ-ViT (Swin-B)

megvii-research/FQ-ViT

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

2021

FQ-ViT (Swin-S)

megvii-research/FQ-ViT

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

2021

FQ-ViT (DeiT-B)

megvii-research/FQ-ViT

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

2021

FQ-ViT (Swin-T)

megvii-research/FQ-ViT

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

2021

FQ-ViT (DeiT-S)

megvii-research/FQ-ViT

HPTQ: Hardware-Friendly Post Training Quantization

2021

Xception W8A8

sony/model_optimization

Learned Step Size Quantization

2019

ADLIK-MO-ResNet50-W4A4

zhutmost/lsq-net hustzxd/LSQuantization

Learned Step Size Quantization

2019

ADLIK-MO-ResNet50-W3A4

zhutmost/lsq-net hustzxd/LSQuantization

HPTQ: Hardware-Friendly Post Training Quantization

2021

EfficientNet-B0 ReLU W8A8

sony/model_optimization

Learned Step Size Quantization

2019

ResNet50-W4A4 (paper)

zhutmost/lsq-net hustzxd/LSQuantization

HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs

2020

EfficientNet-B0-W8A8

sony-si/ai-research sony-si/ai-research

HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs

2020

EfficientNet-B0-W4A4

sony-si/ai-research sony-si/ai-research

HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs

2020

ResNet50-W3A4

sony-si/ai-research sony-si/ai-research

HPTQ: Hardware-Friendly Post Training Quantization

2021

EfficientNet-B0 W8A8

sony/model_optimization

Multi-Prize Lottery Ticket Hypothesis: Finding Accurate Binary Neural Networks by Pruning A Randomly Weighted Network

2021

MPT (80) +BN

chrundle/biprop

LSQ+: Improving low-bit quantization through learnable offsets and better initialization

2020

EfficientNet-W4A4

ZouJiu1/LSQplus yashbhalgat/QualcommAI-MicroNet-submission-MixNet

HPTQ: Hardware-Friendly Post Training Quantization

2021

DenseNet-121 W8A8

sony/model_optimization

LSQ+: Improving low-bit quantization through learnable offsets and better initialization

2020

MixNet-W4A4

ZouJiu1/LSQplus yashbhalgat/QualcommAI-MicroNet-submission-MixNet

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

2021

FQ-ViT (DeiT-T)

megvii-research/FQ-ViT

Training Multi-bit Quantized and Binarized Networks with A Learnable Symmetric Quantizer

2021

UniQ (Ours)

phuocphn/uniq

HPTQ: Hardware-Friendly Post Training Quantization

2021

MobileNetV2 W8A8

sony/model_optimization

HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs

2020

MobileNetV2

sony-si/ai-research sony-si/ai-research

R2 Loss: Range Restriction Loss for Model Compression and Quantization

2023

MobileNet-v1 + EWGS + R2Loss

R2 Loss: Range Restriction Loss for Model Compression and Quantization

2023

MobileNet-v1 + LSQ + R2Loss

R2 Loss: Range Restriction Loss for Model Compression and Quantization

2023

ResNet-18 + PACT + R2Loss

ImageNet

Performance Over Time

Edit Benchmark Results

Edit Result

Top Performing Models

All Papers (27)

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

HPTQ: Hardware-Friendly Post Training Quantization

Learned Step Size Quantization

Learned Step Size Quantization

HPTQ: Hardware-Friendly Post Training Quantization

Learned Step Size Quantization

HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs

HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs

HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs

HPTQ: Hardware-Friendly Post Training Quantization

Multi-Prize Lottery Ticket Hypothesis: Finding Accurate Binary Neural Networks by Pruning A Randomly Weighted Network

LSQ+: Improving low-bit quantization through learnable offsets and better initialization

HPTQ: Hardware-Friendly Post Training Quantization

LSQ+: Improving low-bit quantization through learnable offsets and better initialization

FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer

Training Multi-bit Quantized and Binarized Networks with A Learnable Symmetric Quantizer

HPTQ: Hardware-Friendly Post Training Quantization

HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs

R2 Loss: Range Restriction Loss for Model Compression and Quantization

R2 Loss: Range Restriction Loss for Model Compression and Quantization

R2 Loss: Range Restriction Loss for Model Compression and Quantization

Model	Paper	Top-1 Accuracy (%)	Date
FQ-ViT (ViT-L)	FQ-ViT: Post-Training Quantization for Fully Quan…	85.03	2021-11-27
FQ-ViT (ViT-B)	FQ-ViT: Post-Training Quantization for Fully Quan…	83.31	2021-11-27
FQ-ViT (Swin-B)	FQ-ViT: Post-Training Quantization for Fully Quan…	82.97	2021-11-27
FQ-ViT (Swin-S)	FQ-ViT: Post-Training Quantization for Fully Quan…	82.71	2021-11-27
FQ-ViT (DeiT-B)	FQ-ViT: Post-Training Quantization for Fully Quan…	81.20	2021-11-27
FQ-ViT (Swin-T)	FQ-ViT: Post-Training Quantization for Fully Quan…	80.51	2021-11-27
FQ-ViT (DeiT-S)	FQ-ViT: Post-Training Quantization for Fully Quan…	79.17	2021-11-27
Xception W8A8	HPTQ: Hardware-Friendly Post Training Quantization	78.97	2021-09-19
ADLIK-MO-ResNet50-W4A4	Learned Step Size Quantization	77.88	2019-02-21
ADLIK-MO-ResNet50-W3A4	Learned Step Size Quantization	77.34	2019-02-21
EfficientNet-B0 ReLU W8A8	HPTQ: Hardware-Friendly Post Training Quantization	77.09	2021-09-19
ResNet50-W4A4 (paper)	Learned Step Size Quantization	76.70	2019-02-21
EfficientNet-B0-W8A8	HMQ: Hardware Friendly Mixed Precision Quantizati…	76.40	2020-07-20
EfficientNet-B0-W4A4	HMQ: Hardware Friendly Mixed Precision Quantizati…	76.00	2020-07-20
ResNet50-W3A4	HMQ: Hardware Friendly Mixed Precision Quantizati…	75.45	2020-07-20
EfficientNet-B0 W8A8	HPTQ: Hardware-Friendly Post Training Quantization	74.22	2021-09-19
MPT (80) +BN	Multi-Prize Lottery Ticket Hypothesis: Finding Ac…	74.03	2021-03-17
EfficientNet-W4A4	LSQ+: Improving low-bit quantization through lear…	73.80	2020-04-20
DenseNet-121 W8A8	HPTQ: Hardware-Friendly Post Training Quantization	73.36	2021-09-19
MixNet-W4A4	LSQ+: Improving low-bit quantization through lear…	71.70	2020-04-20
FQ-ViT (DeiT-T)	FQ-ViT: Post-Training Quantization for Fully Quan…	71.61	2021-11-27
UniQ (Ours)	Training Multi-bit Quantized and Binarized Networ…	71.50	2021-04-01
MobileNetV2 W8A8	HPTQ: Hardware-Friendly Post Training Quantization	71.46	2021-09-19
MobileNetV2	HMQ: Hardware Friendly Mixed Precision Quantizati…	70.90	2020-07-20
MobileNet-v1 + EWGS + R2Loss	R2 Loss: Range Restriction Loss for Model Compres…	69.79	2023-03-14
MobileNet-v1 + LSQ + R2Loss	R2 Loss: Range Restriction Loss for Model Compres…	69.64	2023-03-14
ResNet-18 + PACT + R2Loss	R2 Loss: Range Restriction Loss for Model Compres…	68.45	2023-03-14