Quantization on ImageNet

34개 결과 · ⬇ CSV · JSON

Top-1 Accuracy (%)

Rank	Model	Top-1 Accuracy (%)	Weight bits	Activation bits	Paper	Code	Year
1	AQ4SViT 자동 추출	90	–	–	AQ4SViT: An Automated Quantization Framework with Search Gating Policy for Compressing Spiking Vision Transformers		2026
2	FQ-ViT (ViT-L)	85.03	8	8	FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer	megvii-research/FQ-ViT	2021
3	FQ-ViT (ViT-B)	83.31	8	8	FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer	megvii-research/FQ-ViT	2021
4	FQ-ViT (Swin-B)	82.97	8	8	FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer	megvii-research/FQ-ViT	2021
5	FQ-ViT (Swin-S)	82.71	8	8	FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer	megvii-research/FQ-ViT	2021
6	FQ-ViT (DeiT-B)	81.20	8	8	FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer	megvii-research/FQ-ViT	2021
7	FQ-ViT (Swin-T)	80.51	8	8	FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer	megvii-research/FQ-ViT	2021
8	FQ-ViT (DeiT-S)	79.17	8	8	FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer	megvii-research/FQ-ViT	2021
9	Xception W8A8	78.972	8	8	HPTQ: Hardware-Friendly Post Training Quantization	sony/model_optimization	2021
10	ADLIK-MO-ResNet50-W4A4	77.878	4	4	Learned Step Size Quantization	zhutmost/lsq-net · hustzxd/LSQuantization · ZouJiu1/LSQplus · +6	2019
11	ADLIK-MO-ResNet50-W3A4	77.34	3	4	Learned Step Size Quantization	zhutmost/lsq-net · hustzxd/LSQuantization · ZouJiu1/LSQplus · +6	2019
12	EfficientNet-B0 ReLU W8A8	77.092	8	8	HPTQ: Hardware-Friendly Post Training Quantization	sony/model_optimization	2021
13	Dynamic 자동 추출	77.00	–	–	DQT: Dynamic Quantization Training via Dequantization-Free Nested Integer Arithmetic		2025
14	ResNet50-W4A4 (paper)	76.7	4	4	Learned Step Size Quantization	zhutmost/lsq-net · hustzxd/LSQuantization · ZouJiu1/LSQplus · +6	2019
15	EfficientNet-B0-W8A8	76.4	8	8	HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs	sony-si/ai-research · sony-si/ai-research	2020
16	EfficientNet-B0-W4A4	76	4	4	HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs	sony-si/ai-research · sony-si/ai-research	2020
17	ResNet50-W3A4	75.45	3	4	HMQ: Hardware Friendly Mixed Precision Quantization Block for CNNs	sony-si/ai-research · sony-si/ai-research	2020
18	EfficientNet-B0 W8A8	74.216	8	8	HPTQ: Hardware-Friendly Post Training Quantization	sony/model_optimization	2021
19	MPT (80) +BN	74.03	–	–	Multi-Prize Lottery Ticket Hypothesis: Finding Accurate Binary Neural Networks by Pruning A Randomly Weighted Network	chrundle/biprop	2021
20	ProfilingAgent 자동 추출	74	–	–	ProfilingAgent: Profiling-Guided Agentic Reasoning for Adaptive Model Optimization		2025