yihaopeng's picture
dreamstruct-screen-classification init
3156eb6
from torchvision.io import read_image
from torchvision.models import resnet50, ResNet50_Weights
import glob, os, csv
import numpy as np
from PIL import Image
from torchvision import models, transforms
import torch
import torch.optim as optim
import torch.nn as nn
import torchvision.transforms as transforms
from torch.utils.data import DataLoader, random_split
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print(device)
# Custom Dataset Class (Replace with your actual dataset class)
class CustomDataset(torch.utils.data.Dataset):
def __init__(self, root, transform=None):
# Your dataset initialization code here
self.transform = transform
# read ui types in the csv file design_topics.csv
dict_id_to_ui_type = {}
with open('../enrico/design_topics.csv') as csv_file:
csv_reader = csv.reader(csv_file, delimiter=',')
ui_types_set = set()
for row in csv_reader:
ui_types_set.add(row[1])
if row[1] == 'news':
row[1] = 'gallery'
dict_id_to_ui_type[row[0]] = row[1]
ui_types_list = ['list', 'login', 'settings', 'menu', 'mediaplayer', 'form', 'profile', 'gallery']#list(ui_types_set)
path = root
folders = os.listdir(path)
self.image_list = []
self.ui_type_list = []
c = 0
for f in folders:
c += 1
if c % 50 == 0:
print(c)
image_path = path + f
# image = read_image(image_path)
# # resize to 1280 x 720
# image = image.resize((720, 1280))
# open image as numpy array
# image = Image.open(image_path)
# # resize the image to 1280 x 720
# image = image.resize((720, 1280))
# image.save(image_path)
image = read_image(image_path).to(torch.uint8)
self.image_list.append(image)
# get the ui type of the image
# image_name = '_'.join(f.split('_')[:2])
image_id = f.split('.')[0]
ui_type = dict_id_to_ui_type[image_id]
ui_type_index = ui_types_list.index(ui_type)
label = torch.zeros(8).to(device)
label[ui_type_index] = 1
self.ui_type_list.append(label)
def __len__(self):
# Return the number of samples in your dataset
return len(self.image_list)
def __getitem__(self, idx):
# Load and return a sample from your dataset
img = self.image_list[idx]
if self.transform:
img = self.transform(img)
label = self.ui_type_list[idx]
return img, label
#img = read_image("./UI_images/enrico_124.jpg")
# Step 1: Initialize model with the best available weights
#weights = ResNet50_Weights.DEFAULT
#model = resnet50(weights=weights)
weights = ResNet50_Weights.DEFAULT
resnet50 = models.resnet50(pretrained=True)
#for param in resnet50.parameters():
# param.requires_grad = False
num_ftrs = resnet50.fc.in_features
resnet50.fc = nn.Linear(num_ftrs, 8)
#num_features = model.fc.in_features
#model = torch.nn.Sequential(*(list(model.children())[:-1]))
#model.fc = nn.Linear(model.fc.in_features, 8)
#layer = nn.Linear(num_features, 8)
resnet50.to(device)
#layer.to(device)
#for n, p in model.named_parameters():
# p.require_grad = False
#img = read_image("./UI_images/enrico_124.jpg")
# Step 2: Initialize the inference transforms
#preprocess = weights.transforms()
transform = transforms.Compose([
transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])
preprocess = weights.transforms()
train_dataset = CustomDataset(root="./imgs/train/",
transform=transforms.Compose([preprocess]))
test_dataset = CustomDataset(root="./imgs/test/",
transform=transforms.Compose([preprocess]))
# Create data loaders
train_loader = DataLoader(train_dataset, batch_size=256, shuffle=True)
val_loader2 = DataLoader(test_dataset, batch_size=1, shuffle=False)
# Define loss function and optimizer
criterion = torch.nn.CrossEntropyLoss()
criterion = torch.nn.BCEWithLogitsLoss()
other_p = []
fc_p = []
for name, param in resnet50.named_parameters():
if not name.startswith('fc'):
other_p.append(param)
else:
print(name)
fc_p.append(param)
params = [
{'params': fc_p, 'lr': 0.001},
{'params': other_p, 'lr': 0.0001}
]
optimizer = optim.Adam(params, lr=0.001)
#optimizer = optim.Adam(resnet50.parameters(), lr=0.0001)
#optimizer = optim.Adam(resnet50.fc, lr=0.001)
# Training loop
num_epochs = 501
for epoch in range(num_epochs):
print('epoch', epoch)
resnet50.train()
for inputs, labels in train_loader:
inputs = inputs.to(device)
labels = labels.to(device)
optimizer.zero_grad()
outputs = resnet50(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
# compute the accuracy of each class
# compute the accuracy of each class
correct = 0
total = 0
with torch.no_grad():
if epoch != 0 and epoch % 5 == 0:
# accuracy per class
class_correct = list(0. for i in range(8))
class_total = list(0. for i in range(8))
with torch.no_grad():
for inputs, labels in val_loader2:
inputs = inputs.to(device)
labels = labels.to(device)
outputs = resnet50(inputs)
_, predicted = torch.max(outputs, 1)
# c = (predicted == labels).squeeze()
label_index = torch.where(labels[0] == 1)[0].item()
label = int(label_index)
c = (predicted == label).squeeze()
class_correct[label] += c.item()
class_total[label] += 1
correct += c.item()
total += 1
print('Accuracy of the netxork on the test images: %.2f %%' % (
100 * correct / total))
print(class_correct)
print(class_total)
for i in range(8):
print('Accuracy of %5s : %.2f %%' % (
i, 100 * class_correct[i] / class_total[i]))
# Step 4: Use the model and print the predicted category
# prediction = model(128).squeeze(0).softmax(0)
# class_id = prediction.argmax().item()
# score = prediction[class_id].item()
# category_name = weights.meta["categories"][class_id]
# print(f"{category_name}: {100 * score:.1f}%")