How can the EfficientNet model use in multi-labels image classification?

[ljk1072911239]

How can the EfficientNet model use in multi-labels image classification?

[hegc]

Same as the other archs

[ljk1072911239]

Here is sample of how I used it for 5 class classification.

#load model and define number of classes
model = EfficientNet.from_pretrained('efficientnet-b7', num_classes=5)

#Defining Optimizer and Loss Criteria
criterion = nn.CrossEntropyLoss()
learning_rate=1e-3
optimizer = optim.Adam(model.parameters(), lr=learning_rate)

#Trainer 

for epoch in range(epochs):  
    running_loss = 0.0 #loss
    correct=0 #num_correct in a batch
    total=0 #total_size

    for i, data in enumerate(training_generator, 0):
        inputs, labels = data
       
        inputs, labels = inputs.to(device), labels.to(device)
        labels = eye[labels]
        
        #calc loss
        optimizer.zero_grad()
        outputs = model(inputs)
        loss = criterion(outputs, torch.max(labels, 1)[1])
        _, predicted = torch.max(outputs, 1)
        _, labels = torch.max(labels, 1)
        
        correct += (predicted == labels).sum().item()
        total += labels.size(0)
        accuracy = float(correct) / float(total)

        #backward prop
        loss.backward()
        optimizer.step()

what is "training_generator"? And why use "labels = eye[labels]" ? Thank you!

[ljk1072911239]

Same as the other archs

Do you have the example code for multi labels classification? Thank you!

[kartikdutt18]

Here is sample of how I used it for 5 class classification.

#load model and define number of classes
model = EfficientNet.from_pretrained('efficientnet-b7', num_classes=5)

#Defining Optimizer and Loss Criteria
criterion = nn.CrossEntropyLoss()
learning_rate=1e-3
optimizer = optim.Adam(model.parameters(), lr=learning_rate)

#Trainer 

for epoch in range(epochs):  
    running_loss = 0.0 #loss
    correct=0 #num_correct in a batch
    total=0 #total_size

    for i, data in enumerate(training_generator, 0):
        inputs, labels = data
       
        inputs, labels = inputs.to(device), labels.to(device)
        labels = eye[labels]
        
        #calc loss
        optimizer.zero_grad()
        outputs = model(inputs)
        loss = criterion(outputs, torch.max(labels, 1)[1])
        _, predicted = torch.max(outputs, 1)
        _, labels = torch.max(labels, 1)
        
        correct += (predicted == labels).sum().item()
        total += labels.size(0)
        accuracy = float(correct) / float(total)

        #backward prop
        loss.backward()
        optimizer.step()

what is "training_generator"? And why use "labels = eye[labels]" ? Thank you!

Training Generator is Data-Generator for the dataset I was working on and eye generates a tensor of NxN. Here is the full code in my repo: https://github.com/kartikdutt18/Kaggle-ATPOS-Efficient_Net/blob/master/Efficient_Net.ipynb If you have trouble viewing it : Use this link https://www.kaggle.com/sanninjiraya/efficient-net or clone the repo.

[hummingbird1989]

Here is sample of how I used it for 5 class classification.

#load model and define number of classes
model = EfficientNet.from_pretrained('efficientnet-b7', num_classes=5)

#Defining Optimizer and Loss Criteria
criterion = nn.CrossEntropyLoss()
learning_rate=1e-3
optimizer = optim.Adam(model.parameters(), lr=learning_rate)

#Trainer 

for epoch in range(epochs):  
    running_loss = 0.0 #loss
    correct=0 #num_correct in a batch
    total=0 #total_size

    for i, data in enumerate(training_generator, 0):
        inputs, labels = data
       
        inputs, labels = inputs.to(device), labels.to(device)
        labels = eye[labels]
        
        #calc loss
        optimizer.zero_grad()
        outputs = model(inputs)
        loss = criterion(outputs, torch.max(labels, 1)[1])
        _, predicted = torch.max(outputs, 1)
        _, labels = torch.max(labels, 1)
        
        correct += (predicted == labels).sum().item()
        total += labels.size(0)
        accuracy = float(correct) / float(total)

        #backward prop
        loss.backward()
        optimizer.step()

This doesn't answer the original question. You're applying the problem to a multi-class problem whereas a multi-label problem means an input can have multiple labels