Context

This is a Keras image data generator ready version of the Plant Seedlings Dataset of the Aarhus University Department of Engineering Signal Processing Group.

This Dataset was previously used in a Kaggle Competition but has been re-uploaded to make working with the data in Keras easier

Content

The images presented show weed and crop seedlings. Your task is to classify the type of plant by an image of its seedling. The images have already been segmented, so that each image shows only one plant.

Acknowledgements

Big thanks to Aarhus University Department of Engineering Signal Processing Group for publishing the dataset

Introduction

Currently support tensorflow in - ResNeSt 2d&3d - RegNet - DETR (modified classfication) - GENet (2020 GPU-Efficient Network)

model only, no pertrain model for download (simply not enough free time and resource).
easy to read and modified. welcome for using it, ask question, test it, find some bugs maybe.

ResNeSt based on offical github .

Update

2020-7-30: Add and based on GENet: A GPU-Efficient Network. Paper shows achieve good performence under GPU environment, very similar to RegNet. model name GENet_light, GENet_normal, GENet_large.

2020-6-14: Add Resnest3D, thanks to @vitanuan, model name resnest50_3d, resnest101_3d, resnest200_3d, input shape is 4d like input_shape = [50,224,224,3]

2020-6-5: Add DETR (res34, resNest50 backbone) End-to-End Object Detection with Transformers, Experiment and inovation model, i slightly modified it into a classficaiton verison. Free to try.

2020-5-27: ResNeSt add CB-Net style to enahce backbone. theoretically, it should improve the results. Wait for test.

Usage

usage is simple: ``` from models.model_factory import get_model

model_name = 'ResNest50' input_shape = [224,244,3] n_classes = 81 fc_activation = 'softmax' active = 'relu' # relu or mish

model = get_model(model_name=model_name, input_shape=input_shape, n_classes=n_classes, fc_activation=fc_activation, active=active', verbose=False, )

model.compile(optimizer='adam', loss=tf.keras.losses.BinaryCrossentropy()) model.fit(...) ```

• if you add CB_Net in ResNeSt: add using_cb=True like: ``` """ Beware that if using CB_Net, the input height and width should be divisibility of 2 at least 5 times, like input [224,448]: [224,448]<->[112,224]<->[56,112]<->[28,56]<->[14,28]<->[7,14] correct way: [224,224] downsample->[112,112], [112,112] upsample->[224,224], then [224,224]+[224,224]

incorrect way: [223,223] downsample->[112,112], [112,112] upsample->[224,224], [223,223] != [224,224] cannt add """

model = get_model(...,using_cb=True) - DETR experiment model, free to modified the transformer setting.

ResNest50+CB+transfomer looks powerful! but heavily cost.

model_name = 'ResNest50_DETR'

res34 not implement using_cb yet, it supporse to be a lighter verison.

model_name = 'res34_DETR'

model = get_model(..., hidden_dim=512, nheads=8, num_encoder_layers=6, num_decoder_layers=6, n_query_pos=100) ```

Models

models now support: ``` ResNest50 ResNest101 ResNest200 ResNest269

RegNetX400 RegNetX1.6 RegNetY400 RegNetY1.6 AnyOther RegNetX/Y

res34_DETR ResNest50_DETR ```

RegNet

for RegNet, cause there are various version, you can easily set it by stage_depth,stage_width,stage_G.

#RegNetY600
model = get_model(model_name="RegNet",input_shape=input_shape,n_classes=n_classes,
        verbose=True,fc_activation=fc_activation,stage_depth=[1,3,7,4],
        stage_width=[48,112,256,608],stage_G=16,SEstyle_atten="SE")

#RegNetX600
model = get_model(model_name="RegNet",input_shape=input_shape,n_classes=n_classes,
        verbose=True,fc_activation=fc_activation,stage_depth=[1,3,5,7],
        stage_width=[48,96,240,528],stage_G=24,SEstyle_atten="noSE")

details seting (from orginal paper ): - facebookresearch/pycls

https://raw.githubusercontent.com/QiaoranC/tf_ResNeSt_RegNet_model/master/readme_img/regnet_setting.png" alt="alt text">

• CB-Net, using this style to enhace ResNest https://raw.githubusercontent.com/QiaoranC/tf_ResNeSt_RegNet_model/master/readme_img/CBNet.png" alt="alt text">

DETR

• orginal is detection version with two linear out (box and cls), here only clsficaiton (with some modification), if anyone interesting in the detection version, asked and i can change to a deteciton version.
  • torch version using nn.Parameter to buildt q,k,v, but in tensorflow2.x, i tried similar Variable, but the Variable shape can't set batch dimed like (?,100,100), and model layers output are like (?,7,7,2048), So to combine Variable into model, i use a trick way, to make a fake layer out (?,100,100), and add with the Variable (100,100), Variable will become (?,100,100) then can feed into transformer. I didnot found a better way. Again, Welcome any good ideal or suggesions.

Discussion

I compared ResNeSt50 and some RegNet(below 4.0GF) in my own project, also compared to EfficientNet b0/b1/b2. it seems EfficientNet is still go...

BIRDS 20 SPECIES- IMAGE CLASSIFICATION Data set of 20 bird species. 3208 training images, 100 test images(5 images per species) and 100 validation images(5 images per species. This is a very high quality dataset where there is only one bird in each image and the bird typically takes up at least 50% of the pixels in the image. As a result even a moderately complex model will achieve training and test accuracies in the mid 90% range. Note: all images are original and not created by augmentation All images are 224 X 224 X 3 color images in jpg format. Data set includes a train set, test set and validation set. Each set contains 475 sub directories, one for each bird species. The data structure is convenient if you use the Keras ImageDataGenerator.flow_from_directory to create the train, test and valid data generators. The data set also include a file birds.csv. This cvs file contains 5 columns. The filepaths column contains the relative file path to an image file. The labels column contains the bird species class name associated with the image file. The scientific label column contains the latin scientific name for the image. The data set column denotes which dataset (train, test or valid) the filepath resides in. The class_id column contains the class index value associated with the image file's class. NOTE: The test and validation images in the data set were hand selected to be the "best" images so your model will probably get the highest accuracy score using those data sets versus creating your own test and validation sets. However the latter case is more accurate in terms of model performance on unseen images. Images were gather from internet searches by species name. Once the image files for a species was downloaded they were checked for duplicate images using a python duplicate image detector program I developed. All duplicate images detected were deleted in order to prevent their being images common between the training, test and validation sets. After that the images were cropped so that the bird in most cases occupies at least 50% of the pixel in the image. Then the images were resized to 224 X 224 X3 in jpg format. The cropping ensures that when processed by a CNN their is adequate information in the images to create a highly accurate classifier. Even a moderately robust model should achieve training, validation and test accuracies in the high 90% range. Because of the large size of the dataset I recommend if you try to train a model use and image size of 150 X 150 X 3 in order to reduce training time. All files were also numbered sequential starting from one for each species. So test images are named 1.jpg to 5.jpg. Similarly for validation images. Training images are also numbered sequentially with "zeros" padding. For example 001.jpg, 002.jpg ….010.jpg, 011.jpg …..099.jpg, 100jpg, 102.jpg etc. The zero's padding preserves the file order when used with python file functions and Keras flow from directory. The training set is not balanced, having a varying number of files per species. However each species has at least 130 training image files. One significant shortcoming in the data set is the ratio of male species images to female species images. About 80% of the images are of the male and 20% of the female. Males typical are far more diversely colored while the females of a species are typically bland. Consequently male and female images may look entirely different .Almost all test and validation images are taken from the male of the species. Consequently the classifier may not perform as well on female specie images.

The dataset contains images categorized into sehat and tidak sehat , organized into train , test , and validation folders, each with subfolders for each class ( /sehat and /tidak sehat ). Images are in JPEG or PNG format with a recommended resolution of 240x240 pixels, suitable for the VGG16 model’s input requirements. The dataset is intended for deep learning applications, viewable with standard image viewers, and executable with Python, particularly using TensorFlow and Keras . To access and run the VGG16 model, Google Colab or Jupyter Notebook can be used for cloud. For processing, an image data generator is set up to normalize the images, while VGG16 (with pre-trained ImageNet weights) serves as the base model with added dense layers for binary classification between sehat and tidak sehat . The model can then be compiled with an optimizer (e.g., Adam) and trained on the data with appropriate evaluation on validation and test sets.

A spatio-temporal (ST) machine learning (ML) model for security-constrained unit commitment (SCUC) solution acceleration. The ML architecture with GNN and LSTM layers. Includes two models, one for node prediction to predict generator commitment status, and another for edge prediction, which predicts congested lines in the system. The predictions from the two models are then used to reduce the number of variables and constraints in a SCUC problem.NOTE: Codes are implemented in Python. ML model uses Keras, Tensorflow and Spektral (GNN) libraries. Optimization is implemented using Pyomo in python. A solver license (cplex/gurobi) is required for pyomo to run.

This is an augmented version of the original QR-code dataset (https://www.kaggle.com/datasets/coledie/qr-codes). The purpose of this set is to simulate real-life data by adding noise, random cropping, shear, and rotation. This would help create more robust Object-detection and Generative models. In addition, the below code would also help with creating your very own Augmented dataset.

import cv2 
import os
datadir = 'qr_dataset' #you'll have to change datadir accordingly
array = []
array_small =[]
from tqdm import tqdm
def create_training_data():
    for img in tqdm(list(os.listdir(datadir))): # iterate over each image per dogs and cats
      try:
        img_array = cv2.imread(datadir+'/'+img ,cv2.IMREAD_COLOR) # convert to array
        new_array = cv2.resize(img_array, (128, 128)) # resize to normalize data size
        array.append([new_array]) 
        array_small.append([cv2.resize(img_array, (32,32),
                interpolation=cv2.INTER_AREA)]) # add this to our training_data
      except Exception as e: # in the interest in keeping the output clean...
        pass
create_training_data()

#augmenting the data
from keras.preprocessing.image import ImageDataGenerator #this generator will save files in a physical format
from skimage import io
datagen = ImageDataGenerator(    
    rotation_range = 40,
    shear_range = 0.2,
    zoom_range = 0.2,
    horizontal_flip = True,
    brightness_range = (0.5, 1.5))

for a in X:
 i = 0
 a = a.reshape((1, ) + a.shape)
 for batch in datagen.flow(a, batch_size=1, save_to_dir= 'Augmented-images', save_prefix='dr', save_format='jpeg'):  
 try: #iterate over every image and augment it
  i += 1  
  if i>= 10:
   break 
 except Exception: #in case the image doesn't exist
  print("error")
  pass

This data set is a collection of 2,000 Bike and Car images. While collecting these images, It was made sure that all types of bikes and cars are included in the image collection. This is because of the high Intra-variety of cars and bikes. That is, there are different types of cars and bikes, which make it a little tough task for the model because the model will also have to understand the high variety of bikes and cars. But if your model is able to understand the basic structure of a car and a bike, it will be able to distinguish between both classes.

The data is not preprocessed. This is done intentionally so that you can apply the augmentations you want to use. Almost all the 2000 images are unique. So after applying some data augmentation, you can increase the size of the data set.

The data is not distributed into training and validation subsets. But you can easily do so by using an Image data generator from Keras. The preprocessing steps are available in the my notebook associated with this data set. You can practice your computer vision skills using this data set. This is a binary classification task.

Overview

The Human Faces (Real, Sketch, Synthetic) dataset is a curated collection of facial images across three distinct domains — Real, Sketch, and Synthetic. Each category contains images of different individuals, offering diverse visual styles and data distributions. This makes the dataset ideal for cross-domain generalization, synthetic face detection, and domain adaptation research, rather than one-to-one identity mapping.

Dataset Summary

Real Faces : 500 images Synthetic Faces : 500 images Sketch Faces : 188 images Total Images : 1,188 images

Note: Each category contains different individuals, not paired samples.

Technical Details

Image Format : .png, .jpg Image Type : RGB Total Images : 1,188 Ideal For : TensorFlow, PyTorch, OpenCV, Keras, and other ML pipelines

Antara tahun 2023 dan 2030, Pasar Sistem Pengendalian Pembangkit Listrik diperkirakan akan berkembang pada CAGR sebesar 6,75%. Ukuran pasar, pertumbuhan, bagikan

Context

I'm a guitar man, I love music and i'm graduated in Informatic science. I would like to merge my passion and create a dataset that contains the entire text of the song of my favourite singer: the young italian singer Niccolò Moriconi in art Ultimo. I have built the dataset from zero, thanks to some open source python scraping library.

Content

It contains the entire text of the song in his 3 Album.

Acknowledgements

Thanks to Niccolò to be a constant inspiration of mine and to enter always in my heart when I listen his beautiful songs.

Inspiration

I'm working for create a text generator algorithm for create a new song using AI. The technology that i'm using is Python and in particular: - Nltk. - Tensorflow. - Keras. - LSTM (Neural Network).