keras/resnet_vd_34_imagenet

Model Overview

Model Summary

Instantiates the ResNet architecture amended by “bag of tricks” modifications.

Reference

Bag of Tricks for Image Classification with Convolutional Neural Networks

ResNetVd introduces two key modifications to the standard ResNet. First, the initial convolutional layer is replaced by a series of three successive convolutional layers. Second, shortcut connections use an additional pooling operation rather than performing downsampling within the convolutional layers themselves.

Links

• ResNetVD Quickstart Notebook
• ResNet and ResNetVD series Doc
• ResNetVD Model Card
• KerasHub Beginner Guide
• KerasHub Model Publishing Guide

Installation

Keras and KerasHub can be installed with:

pip install -U -q keras-hub
pip install -U -q keras

Jax, TensorFlow, and Torch come preinstalled in Kaggle Notebooks. For instructions on installing them in another environment see the Keras Getting Started page.

Presets

The following model checkpoints are provided by the Keras team.Weights have been ported from: PaddleOCR. Full code examples for each are available below.

Preset name	Parameters	Description
resnet_vd_18_imagenet	11.72M	18-layer ResNetVD model pre-trained on the ImageNet 1k dataset at a 224x224 resolution.
resnet_vd_34_imagenet	21.84M	34-layer ResNet model pre-trained on the ImageNet 1k dataset at a 224x224 resolution.
resnet_vd_50_imagenet	25.63M	50-layer ResNet model pre-trained on the ImageNet 1k dataset at a 224x224 resolution.
resnet_vd_50_ssld_imagenet	25.63M	50-layer ResNet model pre-trained on the ImageNet 1k dataset at a 224x224 resolution with knowledge distillation.
resnet_vd_50_ssld_v2_imagenet	25.63M	50-layer ResNet model pre-trained on the ImageNet 1k dataset at a 224x224 resolution with knowledge distillation and AutoAugment.
resnet_vd_50_ssld_v2_fix_imagenet	25.63M	50-layer ResNet model pre-trained on the ImageNet 1k dataset at a 224x224 resolution with knowledge distillation, AutoAugment and additional fine-tuning of the classification head.
resnet_vd_101_imagenet	44.67M	101-layer ResNet model pre-trained on the ImageNet 1k dataset at a 224x224 resolution.
resnet_vd_101_ssld_imagenet	44.67M	101-layer ResNet model pre-trained on the ImageNet 1k dataset at a 224x224 resolution with knowledge distillation.
resnet_vd_152_imagenet	60.36M	152-layer ResNet model pre-trained on the ImageNet 1k dataset at a 224x224 resolution.
resnet_vd_200_imagenet	74.93M	200-layer ResNet model pre-trained on the ImageNet 1k dataset at a 224x224 resolution.

Example Usage

   
from keras_hub.models import ResNetBackbone
import keras
import numpy as np

input_data = np.ones(shape=(8, 224, 224, 3))

# Pretrained backbone
model = ResNetBackbone.from_preset("resnet_vd_34_imagenet")
output = model(input_data)

# Randomly initialized backbone with a custom config
model = ResNetBackbone(
        input_conv_filters=[32, 32, 64],
        input_conv_kernel_sizes=[3, 3, 3],
        stackwise_num_filters=[64, 128, 256, 512],
        stackwise_num_blocks=[3, 4, 5, 6],
        stackwise_num_strides=[1, 2, 2, 2],
        block_type="bottleneck_block_vd",
    )
output = model(input_data)

Example Usage with Hugging Face URI

   
from keras_hub.models import ResNetBackbone
import keras
import numpy as np

input_data = np.ones(shape=(8, 224, 224, 3))

# Pretrained backbone
model = ResNetBackbone.from_preset("hf://keras/resnet_vd_34_imagenet")
output = model(input_data)

# Randomly initialized backbone with a custom config
model = ResNetBackbone(
        input_conv_filters=[32, 32, 64],
        input_conv_kernel_sizes=[3, 3, 3],
        stackwise_num_filters=[64, 128, 256, 512],
        stackwise_num_blocks=[3, 4, 5, 6],
        stackwise_num_strides=[1, 2, 2, 2],
        block_type="bottleneck_block_vd",
    )
output = model(input_data)