tessl/pypi-keras-nightly

Multi-backend deep learning framework providing a unified API for building and training neural networks across JAX, TensorFlow, PyTorch, and OpenVINO backends

—

Pending

Overview

Eval results

Files

Pre-trained Models

Name: tessl/pypi-keras-nightly
Author: tessl

Extensive collection of pre-trained computer vision models with ImageNet weights for transfer learning, feature extraction, and fine-tuning across various architectures including ResNet, EfficientNet, VGG, MobileNet, and Inception families.

Capabilities

ResNet Models

Residual networks with skip connections for deep architectures.

def ResNet50(include_top=True, weights='imagenet', input_tensor=None, input_shape=None, 
             pooling=None, classes=1000, classifier_activation='softmax', **kwargs):
    """
    ResNet50 architecture.
    
    Args:
        include_top (bool): Whether to include classification head
        weights (str): Pre-trained weights ('imagenet' or None)
        input_tensor: Optional input tensor  
        input_shape (tuple): Input shape for images
        pooling (str): Pooling mode ('avg', 'max', or None)
        classes (int): Number of classes for classification
        classifier_activation (str): Activation for final layer
        
    Returns:
        Model: ResNet50 model
    """

def ResNet101(include_top=True, weights='imagenet', **kwargs): ...
def ResNet152(include_top=True, weights='imagenet', **kwargs): ...
def ResNet50V2(include_top=True, weights='imagenet', **kwargs): ...
def ResNet101V2(include_top=True, weights='imagenet', **kwargs): ...
def ResNet152V2(include_top=True, weights='imagenet', **kwargs): ...

EfficientNet Models

Compound scaled convolutional networks optimized for efficiency and accuracy.

def EfficientNetB0(include_top=True, weights='imagenet', input_tensor=None, **kwargs): ...
def EfficientNetB1(include_top=True, weights='imagenet', **kwargs): ...
def EfficientNetB2(include_top=True, weights='imagenet', **kwargs): ...
def EfficientNetB3(include_top=True, weights='imagenet', **kwargs): ...
def EfficientNetB4(include_top=True, weights='imagenet', **kwargs): ...
def EfficientNetB5(include_top=True, weights='imagenet', **kwargs): ...
def EfficientNetB6(include_top=True, weights='imagenet', **kwargs): ...
def EfficientNetB7(include_top=True, weights='imagenet', **kwargs): ...

def EfficientNetV2B0(include_top=True, weights='imagenet', **kwargs): ...
def EfficientNetV2B1(include_top=True, weights='imagenet', **kwargs): ...
def EfficientNetV2B2(include_top=True, weights='imagenet', **kwargs): ...
def EfficientNetV2B3(include_top=True, weights='imagenet', **kwargs): ...
def EfficientNetV2S(include_top=True, weights='imagenet', **kwargs): ...
def EfficientNetV2M(include_top=True, weights='imagenet', **kwargs): ...
def EfficientNetV2L(include_top=True, weights='imagenet', **kwargs): ...

VGG Models

Visual Geometry Group architectures with small convolution filters.

def VGG16(include_top=True, weights='imagenet', input_tensor=None, **kwargs): ...
def VGG19(include_top=True, weights='imagenet', input_tensor=None, **kwargs): ...

MobileNet Models

Lightweight models optimized for mobile and embedded devices.

def MobileNet(include_top=True, weights='imagenet', input_tensor=None, **kwargs): ...
def MobileNetV2(include_top=True, weights='imagenet', input_tensor=None, **kwargs): ...
def MobileNetV3Small(include_top=True, weights='imagenet', **kwargs): ...
def MobileNetV3Large(include_top=True, weights='imagenet', **kwargs): ...

Other Architectures

def InceptionV3(include_top=True, weights='imagenet', **kwargs): ...
def InceptionResNetV2(include_top=True, weights='imagenet', **kwargs): ...
def Xception(include_top=True, weights='imagenet', **kwargs): ...
def DenseNet121(include_top=True, weights='imagenet', **kwargs): ...
def DenseNet169(include_top=True, weights='imagenet', **kwargs): ...
def DenseNet201(include_top=True, weights='imagenet', **kwargs): ...
def NASNetMobile(include_top=True, weights='imagenet', **kwargs): ...
def NASNetLarge(include_top=True, weights='imagenet', **kwargs): ...

Usage Examples

Transfer Learning

import keras
from keras.applications import ResNet50
from keras import layers

# Load pre-trained model without top classification layer
base_model = ResNet50(weights='imagenet', include_top=False, input_shape=(224, 224, 3))

# Freeze base model weights
base_model.trainable = False

# Add custom classification head
model = keras.Sequential([
    base_model,
    layers.GlobalAveragePooling2D(),
    layers.Dense(128, activation='relu'),
    layers.Dropout(0.2),
    layers.Dense(num_classes, activation='softmax')
])

model.compile(
    optimizer='adam',
    loss='sparse_categorical_crossentropy', 
    metrics=['accuracy']
)

Feature Extraction

from keras.applications import EfficientNetB0
from keras.applications.efficientnet import preprocess_input

# Load model for feature extraction
model = EfficientNetB0(weights='imagenet', include_top=False, pooling='avg')

# Preprocess images
x = preprocess_input(images)

# Extract features
features = model.predict(x)

Install with Tessl CLI