Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -6,22 +6,95 @@ import requests
|
|
6 |
from PIL import Image
|
7 |
from torchvision import transforms
|
8 |
import urllib.request
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
-
#
|
11 |
-
|
12 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
|
14 |
-
# Load the model and tokenizer from your Hugging Face repository
|
15 |
-
model = AutoModel.from_pretrained("Maverick98/EcommerceClassifier")
|
16 |
-
tokenizer = AutoTokenizer.from_pretrained("jinaai/jina-embeddings-v2-base-en")
|
17 |
|
18 |
-
#
|
19 |
transform = transforms.Compose([
|
20 |
transforms.Resize((224, 224)),
|
|
|
|
|
|
|
21 |
transforms.ToTensor(),
|
22 |
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
|
23 |
])
|
24 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
def load_image(image_path_or_url):
|
26 |
"""
|
27 |
Load an image from a URL or local path and preprocess it.
|
@@ -45,7 +118,7 @@ def predict(image_path_or_url, title, threshold=0.7):
|
|
45 |
image = load_image(image_path_or_url)
|
46 |
|
47 |
# Tokenize the title
|
48 |
-
title_encoding = tokenizer(title, padding='max_length', max_length=
|
49 |
input_ids = title_encoding['input_ids']
|
50 |
attention_mask = title_encoding['attention_mask']
|
51 |
|
|
|
6 |
from PIL import Image
|
7 |
from torchvision import transforms
|
8 |
import urllib.request
|
9 |
+
import os
|
10 |
+
import torch
|
11 |
+
import torch.nn as nn
|
12 |
+
import torch.optim as optim
|
13 |
+
from torch.utils.data import DataLoader, Dataset, DistributedSampler
|
14 |
+
from transformers import AutoModel, AutoTokenizer
|
15 |
+
from torchvision import models, transforms
|
16 |
+
from sklearn.model_selection import train_test_split
|
17 |
+
from sklearn.utils.class_weight import compute_class_weight
|
18 |
+
from sklearn.metrics import precision_recall_fscore_support, accuracy_score
|
19 |
+
from torch.cuda.amp import GradScaler, autocast
|
20 |
+
import numpy as np
|
21 |
+
import torch.multiprocessing as mp
|
22 |
+
import torch.distributed as dist
|
23 |
+
import matplotlib.pyplot as plt
|
24 |
|
25 |
+
# --- Define the Model ---
|
26 |
+
class FineGrainedClassifier(nn.Module):
|
27 |
+
def __init__(self, num_classes=434): # Updated to 434 classes
|
28 |
+
super(FineGrainedClassifier, self).__init__()
|
29 |
+
self.image_encoder = models.resnet50(pretrained=True)
|
30 |
+
self.image_encoder.fc = nn.Identity()
|
31 |
+
self.text_encoder = AutoModel.from_pretrained('jinaai/jina-embeddings-v2-base-en')
|
32 |
+
self.classifier = nn.Sequential(
|
33 |
+
nn.Linear(2048 + 768, 1024),
|
34 |
+
nn.BatchNorm1d(1024),
|
35 |
+
nn.ReLU(),
|
36 |
+
nn.Dropout(0.3),
|
37 |
+
nn.Linear(1024, 512),
|
38 |
+
nn.BatchNorm1d(512),
|
39 |
+
nn.ReLU(),
|
40 |
+
nn.Dropout(0.3),
|
41 |
+
nn.Linear(512, num_classes) # Updated to 434 classes
|
42 |
+
)
|
43 |
+
|
44 |
+
def forward(self, image, input_ids, attention_mask):
|
45 |
+
image_features = self.image_encoder(image)
|
46 |
+
text_output = self.text_encoder(input_ids=input_ids, attention_mask=attention_mask)
|
47 |
+
text_features = text_output.last_hidden_state[:, 0, :]
|
48 |
+
combined_features = torch.cat((image_features, text_features), dim=1)
|
49 |
+
output = self.classifier(combined_features)
|
50 |
+
return output
|
51 |
|
|
|
|
|
|
|
52 |
|
53 |
+
# --- Data Augmentation Setup ---
|
54 |
transform = transforms.Compose([
|
55 |
transforms.Resize((224, 224)),
|
56 |
+
transforms.RandomHorizontalFlip(),
|
57 |
+
transforms.RandomRotation(15),
|
58 |
+
transforms.ColorJitter(brightness=0.2, contrast=0.2, saturation=0.2, hue=0.2),
|
59 |
transforms.ToTensor(),
|
60 |
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
|
61 |
])
|
62 |
|
63 |
+
# def load_model_checkpoint(model, checkpoint_path, device):
|
64 |
+
# checkpoint = torch.load(checkpoint_path, map_location=device)
|
65 |
+
|
66 |
+
# # Strip the "module." prefix from the keys in the state_dict if they exist
|
67 |
+
# state_dict = checkpoint['model_state_dict']
|
68 |
+
# new_state_dict = {}
|
69 |
+
|
70 |
+
# for k, v in state_dict.items():
|
71 |
+
# if k.startswith("module."):
|
72 |
+
# new_state_dict[k[7:]] = v # Remove "module." prefix
|
73 |
+
# else:
|
74 |
+
# new_state_dict[k] = v
|
75 |
+
|
76 |
+
# model.load_state_dict(new_state_dict)
|
77 |
+
# return model
|
78 |
+
|
79 |
+
# Load the label-to-class mapping from your Hugging Face repository
|
80 |
+
label_map_url = "https://huggingface.co/Maverick98/EcommerceClassifier/resolve/main/label_to_class.json"
|
81 |
+
label_to_class = requests.get(label_map_url).json()
|
82 |
+
|
83 |
+
|
84 |
+
# Load your custom model from Hugging Face
|
85 |
+
model = FineGrainedClassifier(num_classes=len(label_to_class))
|
86 |
+
model_checkpoint = "Maverick98/EcommerceClassifier"
|
87 |
+
model.load_state_dict(torch.hub.load_state_dict_from_url(f"https://huggingface.co/{model_checkpoint}/resolve/main/model_checkpoint.pth", map_location=torch.device('cpu')))
|
88 |
+
# Load the tokenizer from Jina
|
89 |
+
tokenizer = AutoTokenizer.from_pretrained("jinaai/jina-embeddings-v2-base-en")
|
90 |
+
|
91 |
+
# # Define image preprocessing
|
92 |
+
# transform = transforms.Compose([
|
93 |
+
# transforms.Resize((224, 224)),
|
94 |
+
# transforms.ToTensor(),
|
95 |
+
# transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
|
96 |
+
# ])
|
97 |
+
|
98 |
def load_image(image_path_or_url):
|
99 |
"""
|
100 |
Load an image from a URL or local path and preprocess it.
|
|
|
118 |
image = load_image(image_path_or_url)
|
119 |
|
120 |
# Tokenize the title
|
121 |
+
title_encoding = tokenizer(title, padding='max_length', max_length=200, truncation=True, return_tensors='pt')
|
122 |
input_ids = title_encoding['input_ids']
|
123 |
attention_mask = title_encoding['attention_mask']
|
124 |
|