python 15 lines · 1 tab

Transfer learning with pretrained torchvision backbones

Dr. Elena Vasquez Apr 2026

1 tab

import torch.nn as nn
from torchvision.models import resnet50, ResNet50_Weights

model = resnet50(weights=ResNet50_Weights.IMAGENET1K_V2)

for parameter in model.parameters():
    parameter.requires_grad = False

model.fc = nn.Sequential(
    nn.Dropout(0.3),
    nn.Linear(model.fc.in_features, 4),
)

for parameter in model.fc.parameters():
    parameter.requires_grad = True

1 file · python Explain with highlit

Transfer learning is the right default when labeled data is limited and time matters. I usually freeze the backbone first, train the head, then selectively unfreeze deeper layers if the domain gap justifies it. This strategy converges faster and is much less brittle than training from scratch.

Related snips

python

import torch.nn as nn

class SmallCNN(nn.Module):
    def __init__(self, num_classes: int) -> None:
        super().__init__()
        self.features = nn.Sequential(

Convolutional neural networks for image classification in PyTorch

pytorch cnn computer-vision

by Dr. Elena Vasquez 1 tab

python

import pandas as pd
import torch
from PIL import Image
from torch.utils.data import Dataset, DataLoader

class ProductImageDataset(Dataset):

Custom Datasets and DataLoaders for robust training input pipelines

pytorch dataloader dataset

by Dr. Elena Vasquez 1 tab

python

import torch

best_val_loss = float('inf')

for epoch in range(1, num_epochs + 1):
    model.train()

A clean PyTorch training loop with validation and checkpoints

pytorch training-loop checkpoints

by Dr. Elena Vasquez 1 tab

python

import torch

device = 'cuda' if torch.cuda.is_available() else 'cpu'
features = torch.tensor([[1.0, 2.0], [3.0, 4.0]], requires_grad=True, device=device)
weights = torch.tensor([[0.2], [0.8]], requires_grad=True, device=device)

PyTorch tensor basics and automatic differentiation

pytorch tensors autograd

by Dr. Elena Vasquez 1 tab

Share this code

Here's the card — post it anywhere.

Transfer learning with pretrained torchvision backbones — share card

Link copied