Test Model

🚀 ResNet50 v1.5

A modified version of the ResNet50 v1 model for image classification with improved accuracy.

🚀 Quick Start

The ResNet50 v1.5 model is a modified version of the original ResNet50 v1 model. It offers slightly higher accuracy (~0.5% top1) compared to v1, though with a small performance drawback (~5% imgs/sec).

✨ Features

• Modified Architecture: In the bottleneck blocks requiring downsampling, v1.5 has stride = 2 in the 3x3 convolution, unlike v1 which has it in the first 1x1 convolution.
• Mixed Precision Training: Trained with mixed precision using Tensor Cores on Volta, Turing, and NVIDIA Ampere GPU architectures, enabling over 2x faster results than non - Tensor Core training.
• Deployment Flexibility: Can be deployed for inference on the NVIDIA Triton Inference Server using TorchScript, ONNX Runtime or TensorRT as an execution backend.

📦 Installation

To run the example, you need to install some extra Python packages for preprocessing images and visualization:

!pip install validators matplotlib

💻 Usage Examples

Basic Usage

import torch
from PIL import Image
import torchvision.transforms as transforms
import numpy as np
import json
import requests
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings('ignore')
%matplotlib inline

device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
print(f'Using {device} for inference')

resnet50 = torch.hub.load('NVIDIA/DeepLearningExamples:torchhub', 'nvidia_resnet50', pretrained=True)
utils = torch.hub.load('NVIDIA/DeepLearningExamples:torchhub', 'nvidia_convnets_processing_utils')

resnet50.eval().to(device)

uris = [
    'http://images.cocodataset.org/test-stuff2017/000000024309.jpg',
    'http://images.cocodataset.org/test-stuff2017/000000028117.jpg',
    'http://images.cocodataset.org/test-stuff2017/000000006149.jpg',
    'http://images.cocodataset.org/test-stuff2017/000000004954.jpg',
]

batch = torch.cat(
    [utils.prepare_input_from_uri(uri) for uri in uris]
).to(device)

with torch.no_grad():
    output = torch.nn.functional.softmax(resnet50(batch), dim=1)
    
results = utils.pick_n_best(predictions=output, n=5)

for uri, result in zip(uris, results):
    img = Image.open(requests.get(uri, stream=True).raw)
    img.thumbnail((256,256), Image.ANTIALIAS)
    plt.imshow(img)
    plt.show()
    print(result)

📚 Documentation

For detailed information on model input and output, training recipes, inference and performance visit: github and/or NGC

📄 License

This project is licensed under the Apache-2.0 license.

📖 References

• Original ResNet50 v1 paper
• Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification
• model on github
• model on NGC
• pretrained model on NGC