File size: 5,463 Bytes
fd7d432 b9ca3d0 fd7d432 b9ca3d0 fd7d432 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 |
import torch
import torch.nn as nn
from torchvision import transforms, models
from huggingface_hub import hf_hub_download
from PIL import Image
import torch.nn.functional as F
from MTL import MTL
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])
])
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
def load_model_resnet50_semart():
num_classes_school = 26
num_classes_type = 10
model_path = hf_hub_download(
repo_id="Irina1402/resnet50-semart",
filename="model.pth",
cache_dir="/tmp"
)
# Load the updated MTL model
model = MTL(num_classes_school, num_classes_type)
model.load_state_dict(torch.load(model_path, map_location=device))
model = model.to(device)
model.eval()
school_labels = sorted([
"American", "Austrian", "Belgian", "Bohemian", "Catalan", "Danish", "Dutch", "English", "Finnish",
"Flemish", "French", "German", "Greek", "Hungarian", "Irish", "Italian", "Netherlandish", "Norwegian",
"Other", "Polish", "Portuguese", "Russian", "Scottish", "Spanish", "Swedish", "Swiss"
])
type_labels = sorted([
"genre", "historical", "interior", "landscape", "mythological", "other",
"portrait", "religious", "still-life", "study"
])
return {
"model": model,
"school_labels": school_labels,
"type_labels": type_labels,
"num_classes_school": num_classes_school
}
def load_model_resnet50_balanced():
num_classes_school = 8
num_classes_type = 8
model_path = hf_hub_download(
repo_id="Irina1402/resnet50-balanced",
filename="model.pth",
cache_dir="/tmp"
)
model = models.resnet50(weights=models.ResNet50_Weights.DEFAULT)
num_features = model.fc.in_features
model.fc = nn.Sequential(
nn.Linear(num_features, 512),
nn.ReLU(),
nn.Dropout(0.4),
nn.Linear(512, num_classes_school + num_classes_type)
)
model.load_state_dict(torch.load(model_path, map_location=device))
model = model.to(device)
model.eval()
school_labels = sorted([
"French", "American", "Russian", "British", "Italian", "Spanish", "German", "Dutch"
])
type_labels = sorted([
"portrait", "landscape", "abstract", "genre painting", "religious painting",
"cityscape", "sketch and study", "still life"
])
return {
"model": model,
"school_labels": school_labels,
"type_labels": type_labels,
"num_classes_school": num_classes_school
}
models_registry = {
"model_semart": load_model_resnet50_semart(),
"model_balanced": load_model_resnet50_balanced()
}
def classify_image(image: Image.Image, model_name, confidence_threshold=0.20, strong_threshold=0.80, topk=3):
if model_name not in models_registry:
return {"error": f"Modelul '{model_name}' nu este disponibil."}
model_data = models_registry[model_name]
model = model_data["model"]
school_labels = model_data["school_labels"]
type_labels = model_data["type_labels"]
num_classes_school = model_data["num_classes_school"]
input_tensor = transform(image).unsqueeze(0).to(device)
with torch.no_grad():
if model_name == "model_semart":
# Two outputs directly
school_output, type_output = model(input_tensor)
else:
# Single output to split manually
output = model(input_tensor)
school_output = output[:, :num_classes_school]
type_output = output[:, num_classes_school:]
school_probs = F.softmax(school_output, dim=1).squeeze()
type_probs = F.softmax(type_output, dim=1).squeeze()
# SCHOOL
topk_school = torch.topk(school_probs, k=topk)
school_top1_idx = topk_school.indices[0].item()
school_top1_prob = school_probs[school_top1_idx].item()
if school_top1_prob >= strong_threshold:
school_predictions = [{
"label": school_labels[school_top1_idx],
"score": round(school_top1_prob * 100, 1) # procent
}]
else:
school_predictions = [
{
"label": school_labels[i.item()],
"score": round(school_probs[i].item() * 100, 1)
}
for i in topk_school.indices
if school_probs[i].item() >= confidence_threshold
]
if not school_predictions:
school_predictions = [{"label": "Unknown", "score": None}]
# TYPE
topk_type = torch.topk(type_probs, k=topk)
type_top1_idx = topk_type.indices[0].item()
type_top1_prob = type_probs[type_top1_idx].item()
if type_top1_prob >= strong_threshold:
type_predictions = [{
"label": type_labels[type_top1_idx],
"score": round(type_top1_prob * 100, 1)
}]
else:
type_predictions = [
{
"label": type_labels[i.item()],
"score": round(type_probs[i].item() * 100, 1)
}
for i in topk_type.indices
if type_probs[i].item() >= confidence_threshold
]
if not type_predictions:
type_predictions = [{"label": "Unknown", "score": None}]
return {
"school_prediction": school_predictions,
"type_prediction": type_predictions
}
|