sotabench: deepparrot/EfficientNet-PyTorch (PartyParrot)

lukemelas / EfficientNet-PyTorch

run from github.com/deepparrot/EfficientNet-PyTorch ( forked from github.com/lukemelas/EfficientNet-PyTorch)

Image Classification on ImageNet

	TOP 1 ACCURACY			TOP 5 ACCURACY			SPEED
MODEL	CODE	PAPER	ε-REPR	CODE	PAPER	ε-REPR		PAPER	GLOBAL RANK
EfficientNet-B0	76.8%	76.3%		93.2%	93.2%		314.4		#206
EfficientNet-B1	78.7%	78.8%		94.2%	94.4%		352.1		#136
EfficientNet-B2	79.8%	79.8%		94.7%	94.9%		346.5		#85
EfficientNet-B3	81.1%	81.1%		95.3%	95.5%		304.1		#56
EfficientNet-B4	82.3%	82.6%		96.0%	96.3%		229.6		#37
EfficientNet-B5	83.5%	83.3%		96.7%	96.7%		119.9		#25
EfficientNet-B6	83.9%	84.0%		96.8%	96.9%		69.9		#20
EfficientNet-B7	84.1%	84.4%		96.7%	97.1%		40.4		#17

See Full Build Details

+get badge code

[![SotaBench](https://img.shields.io/endpoint.svg?url=https://sotabench.com/api/v0/badge/gh/deepparrot/EfficientNet-PyTorch)](https://sotabench.com/user/PartyParrot/repos/deepparrot/EfficientNet-PyTorch)

How the Repository is Evaluated

The full sotabench.py file - source

from efficientnet_pytorch.model import EfficientNet
from torchbench.image_classification import ImageNet
import torchvision.transforms as transforms
import PIL

# Define Transforms    
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
b0_input_transform = transforms.Compose([
    transforms.Resize(256, PIL.Image.BICUBIC),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    normalize,
])

# Run Evaluation
ImageNet.benchmark(
    model=EfficientNet.from_pretrained(model_name='efficientnet-b0'),
    paper_model_name='EfficientNet-B0',
    paper_arxiv_id='1905.11946',
    paper_pwc_id='efficientnet-rethinking-model-scaling-for',
    input_transform=b0_input_transform,
    batch_size=256,
    num_gpu=1
)

# Define Transforms    
b1_input_transform = transforms.Compose([
    transforms.Resize(273, PIL.Image.BICUBIC),
    transforms.CenterCrop(240),
    transforms.ToTensor(),
    normalize,
])

# Run Evaluation
ImageNet.benchmark(
    model=EfficientNet.from_pretrained(model_name='efficientnet-b1'),
    paper_model_name='EfficientNet-B1',
    paper_arxiv_id='1905.11946',
    paper_pwc_id='efficientnet-rethinking-model-scaling-for',
    input_transform=b1_input_transform,
    batch_size=256,
    num_gpu=1
)

# Define Transforms    
b2_input_transform = transforms.Compose([
    transforms.Resize(292, PIL.Image.BICUBIC),
    transforms.CenterCrop(260),
    transforms.ToTensor(),
    normalize,
])

# Run Evaluation
ImageNet.benchmark(
    model=EfficientNet.from_pretrained(model_name='efficientnet-b2'),
    paper_model_name='EfficientNet-B2',
    paper_arxiv_id='1905.11946',
    paper_pwc_id='efficientnet-rethinking-model-scaling-for',
    input_transform=b2_input_transform,
    batch_size=256,
    num_gpu=1
)

# Define Transforms    
b3_input_transform = transforms.Compose([
    transforms.Resize(332, PIL.Image.BICUBIC),
    transforms.CenterCrop(300),
    transforms.ToTensor(),
    normalize,
])

# Run Evaluation
ImageNet.benchmark(
    model=EfficientNet.from_pretrained(model_name='efficientnet-b3'),
    paper_model_name='EfficientNet-B3',
    paper_arxiv_id='1905.11946',
    paper_pwc_id='efficientnet-rethinking-model-scaling-for',
    input_transform=b3_input_transform,
    batch_size=256,
    num_gpu=1
)

# Efficient Net B4

# Define Transforms    
b4_input_transform = transforms.Compose([
    transforms.Resize(427, PIL.Image.BICUBIC),
    transforms.CenterCrop(380),
    transforms.ToTensor(),
    normalize,
])

# Run Evaluation
ImageNet.benchmark(
    model=EfficientNet.from_pretrained(model_name='efficientnet-b4'),
    paper_model_name='EfficientNet-B4',
    paper_arxiv_id='1905.11946',
    paper_pwc_id='efficientnet-rethinking-model-scaling-for',
    input_transform=b4_input_transform,
    batch_size=128,
    num_gpu=1
)


# Efficient Net B5

# Define Transforms    
b5_input_transform = transforms.Compose([
    transforms.Resize(512, PIL.Image.BICUBIC),
    transforms.CenterCrop(456),
    transforms.ToTensor(),
    normalize,
])

# Run Evaluation
ImageNet.benchmark(
    model=EfficientNet.from_pretrained(model_name='efficientnet-b5'),
    paper_model_name='EfficientNet-B5',
    paper_arxiv_id='1905.11946',
    paper_pwc_id='efficientnet-rethinking-model-scaling-for',
    input_transform=b5_input_transform,
    batch_size=64,
    num_gpu=1
)

# Efficient Net B6

# Define Transforms    
b6_input_transform = transforms.Compose([
    transforms.Resize(593, PIL.Image.BICUBIC),
    transforms.CenterCrop(528),
    transforms.ToTensor(),
    normalize,
])

# Run Evaluation
ImageNet.benchmark(
    model=EfficientNet.from_pretrained(model_name='efficientnet-b6'),
    paper_model_name='EfficientNet-B6',
    paper_arxiv_id='1905.11946',
    paper_pwc_id='efficientnet-rethinking-model-scaling-for',
    input_transform=b6_input_transform,
    batch_size=64,
    num_gpu=1
)

# Efficient Net B7

# Define Transforms    
b7_input_transform = transforms.Compose([
    transforms.Resize(674, PIL.Image.BICUBIC),
    transforms.CenterCrop(600),
    transforms.ToTensor(),
    normalize,
])

# Run Evaluation
ImageNet.benchmark(
    model=EfficientNet.from_pretrained(model_name='efficientnet-b7'),
    paper_model_name='EfficientNet-B7',
    paper_arxiv_id='1905.11946',
    paper_pwc_id='efficientnet-rethinking-model-scaling-for',
    input_transform=b7_input_transform,
    batch_size=32,
    num_gpu=1
)

STATUS

BUILD

COMMIT MESSAGE

RUN TIME

succeeded

# 5

0h:11m:38s

succeeded

# 4

1h:00m:16s

succeeded

# 3