Some general steps to train a multi-label classification model where each label is a multi-class:

ONE Data Preparation: Prepare your dataset with samples and their corresponding labels, where each label can take on multiple classes. You can represent your labels as binary vectors, where each element in the vector corresponds to a class and is either 1 or 0, depending on whether the sample belongs to that class or not.

TWO Model Selection: Choose a model architecture that is suitable for multi-label classification, such as a neural network with multiple output nodes or a tree-based model with multiple leaves.

THREE Loss Function Selection: Since each label can take on multiple classes, you need to use a loss function that can handle multi-class labels. Some common loss functions for multi-class classification include categorical cross-entropy or focal loss.

FOUR Training: Train your model using the selected loss function, and monitor the performance of your model on both the training and validation sets. You can also use techniques such as regularization or data augmentation to improve the performance of your model.

FIVE Evaluation: Evaluate your model on a test set, and report performance metrics such as accuracy, precision, recall, and F1 score. You can also analyze the performance of your model on each label separately, to see if there are any particular labels that your model is struggling with.

SIX Also, Cross-validation can be utilized in the training step of the multi-label classification where each label is as a multi-class to evaluate the performance of the model and select the best hyperparameters.

SEVEN Fine-tuning: If your model is not performing well on certain labels, you can fine-tune your model by adjusting the hyperparameters, changing the model architecture, or collecting more data for those particular labels.

Multilabeled multiclass classification can be challenging with trial and error making going back and forth between steps.

certain labels, you can fine-tune your model by adjusting the hyperparameters, changing the model architecture, or collecting more data for those particular labels.

Multilabeled multiclass classification can be challenging with trial and error making going back and forth between steps.

If you want to build a more complex classifier that can handle multiple labels for a piece of cloth, you can use a multi-classifier approach. In this approach, you would build a classifier that can first predict the type of label (e.g., color or size), and then use a separate classifier to predict the specific label class.

For example, you could build a classifier that first predicts whether the label for a given piece of cloth is color or size. Once the type of label is predicted, the output of the first classifier could be used as input to a second classifier that predicts the specific label class. This two-step approach can be more efficient and accurate than trying to predict all label classes at once.

Alternatively, you could use a multi-label classifier that can predict multiple labels for a given piece of cloth. In this approach, you would train a single classifier that can predict both the color and size labels for a given piece of cloth, rather than using separate classifiers for each label type.

Overall, the choice of approach depends on the specific requirements and constraints of the task, as well as the availability of labeled data. Multi-classifier and multi-label classifier approaches are both viable options for building a more complex classifier that can handle multiple labels for a piece of cloth.

Multi-class labels are equivalent to multiple binary labels of several classes.

The problem you are trying to tackle is called soft classification (in contrast with hard classification, where one needs to map on object to exclusively one class). Here, we need to map one object to several classes.

The easiest way to do that you implement a set of binary classifiers, each handling one particular one-class label.

Talking about technical details: you need to prepare you data accordingly: you need to transform your multi-class labels into multiple binary one-class labels. Then you need to train multiple independent binary classifiers.

Note here, that some model types (e.g., artificial neural networks with sigmoid output activation) are capable of handling multiple independent binary classes.

Alternatively, you may train three independent hard multi-class classifiers (one for color, one for size and one sleeve type).

can use vgg net model or mask rcnn model