This repository contains the implementation of the Universal Consensus project, a decentralized data exchange protocol that enables secure, transparent, and efficient data sharing and monetization.
To install requirements:
1. pip install -r requirements.txt
- Python 3.6
- GPU Memory: 10GB
- Pytorch 1.4.0
- Getting Started
- Download the dataset: Office-31, OfficeHome, VisDA, DomainNet.
Your dataset DIR:
|-Office/domain_adaptation_images
| |-amazon
| |-webcam
| |-dslr
|-OfficeHome
| |-Art
| |-Product
| |-...
|-VisDA
| |-train
| |-validation
|-DomainNet
| |-clipart
| |-painting
| |-...
You need to modify the data_path in config files, i.e., config.root
Train on one transfer of Office:
1. CUDA_VISIBLE_DEVICES=0 python office_run.py note=EXP_NAME setting=uda/osda/pda source=amazon target=dslr
To train on six transfers of Office:
1. CUDA_VISIBLE_DEVICES=0 python office_run.py note=EXP_NAME setting=uda/osda/pda transfer_all=1
Train on OfficeHome:
1. CUDA_VISIBLE_DEVICES=0 python office_run.py note=EXP_NAME setting=uda/osda/pda source=Art target=Product
office_run.py
1. import torch
2. import torch.nn as nn
3. import torch.optim as optim
4. from torch.utils.data import DataLoader
5. from torchvision import datasets, transforms
6.
7. # Define the model, loss function, and optimizer
8. model = nn.Sequential(...)
9. criterion = nn.CrossEntropyLoss()
10. optimizer = optim.SGD(model.parameters(), lr=0.001)
11.
12. # Load the dataset
13. transform = transforms.Compose([transforms.ToTensor()])
14. dataset = datasets.ImageFolder('path/to/dataset', transform=transform)
15. dataloader = DataLoader(dataset, batch_size=32, shuffle=True)
16.
17. # Train the model
18. for epoch in range(10):
19. for batch in dataloader:
20. inputs, labels = batch
21. optimizer.zero_grad()
22. outputs = model(inputs)
23. loss = criterion(outputs, labels)
24. loss.backward()
25. optimizer.step()
26. print(f'Epoch {epoch+1}, Loss: {loss.item()}')config.json
1. {
2. "root": "path/to/dataset",
3. "uda": {
4. "source": "amazon",
5. "target": "dslr"
6. },
7. "osda": {
8. "source": "Art",
9. "target": "Product"
10. },
11. "pda": {
12. "source": "clipart",
13. "target": "painting"
14. }
15. }Please note that this is just a basic implementation, and you may need to modify the code to suit your specific requirements.