Spaces:
Running
Running
update 2
Browse files
app.py
CHANGED
@@ -6,15 +6,8 @@ import requests
|
|
6 |
from PIL import Image
|
7 |
from torchvision import transforms
|
8 |
import urllib.request
|
9 |
-
import
|
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 torch.cuda.amp import GradScaler, autocast
|
17 |
-
import numpy as np
|
18 |
|
19 |
# --- Define the Model ---
|
20 |
class FineGrainedClassifier(nn.Module):
|
@@ -43,7 +36,6 @@ class FineGrainedClassifier(nn.Module):
|
|
43 |
output = self.classifier(combined_features)
|
44 |
return output
|
45 |
|
46 |
-
|
47 |
# --- Data Augmentation Setup ---
|
48 |
transform = transforms.Compose([
|
49 |
transforms.Resize((224, 224)),
|
@@ -54,60 +46,43 @@ transform = transforms.Compose([
|
|
54 |
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
|
55 |
])
|
56 |
|
57 |
-
# def load_model_checkpoint(model, checkpoint_path, device):
|
58 |
-
# checkpoint = torch.load(checkpoint_path, map_location=device)
|
59 |
-
|
60 |
-
# # Strip the "module." prefix from the keys in the state_dict if they exist
|
61 |
-
# state_dict = checkpoint['model_state_dict']
|
62 |
-
# new_state_dict = {}
|
63 |
-
|
64 |
-
# for k, v in state_dict.items():
|
65 |
-
# if k.startswith("module."):
|
66 |
-
# new_state_dict[k[7:]] = v # Remove "module." prefix
|
67 |
-
# else:
|
68 |
-
# new_state_dict[k] = v
|
69 |
-
|
70 |
-
# model.load_state_dict(new_state_dict)
|
71 |
-
# return model
|
72 |
-
|
73 |
# Load the label-to-class mapping from your Hugging Face repository
|
74 |
label_map_url = "https://huggingface.co/Maverick98/EcommerceClassifier/resolve/main/label_to_class.json"
|
75 |
label_to_class = requests.get(label_map_url).json()
|
76 |
|
77 |
-
|
78 |
# Load your custom model from Hugging Face
|
79 |
model = FineGrainedClassifier(num_classes=len(label_to_class))
|
80 |
-
model_checkpoint = "Maverick98/EcommerceClassifier"
|
81 |
checkpoint_url = f"https://huggingface.co/Maverick98/EcommerceClassifier/resolve/main/model_checkpoint.pth"
|
82 |
checkpoint = torch.hub.load_state_dict_from_url(checkpoint_url, map_location=torch.device('cpu'))
|
83 |
-
|
84 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
85 |
|
86 |
# Load the tokenizer from Jina
|
87 |
tokenizer = AutoTokenizer.from_pretrained("jinaai/jina-embeddings-v2-base-en")
|
88 |
|
89 |
-
|
90 |
-
def load_image(image_path_or_url):
|
91 |
"""
|
92 |
-
|
93 |
"""
|
94 |
-
if image_path_or_url.startswith("http"):
|
95 |
-
with urllib.request.urlopen(image_path_or_url) as url:
|
96 |
-
image = Image.open(url).convert('RGB')
|
97 |
-
else:
|
98 |
-
image = Image.open(image_path_or_url).convert('RGB')
|
99 |
-
|
100 |
image = transform(image)
|
101 |
image = image.unsqueeze(0) # Add batch dimension
|
102 |
return image
|
103 |
|
104 |
-
def predict(
|
105 |
"""
|
106 |
Predict the top 3 categories for the given image and title.
|
107 |
Includes "Others" if the confidence of the top prediction is below the threshold.
|
108 |
"""
|
109 |
# Preprocess the image
|
110 |
-
image = load_image(
|
111 |
|
112 |
# Tokenize the title
|
113 |
title_encoding = tokenizer(title, padding='max_length', max_length=200, truncation=True, return_tensors='pt')
|
@@ -138,7 +113,7 @@ def predict(image_path_or_url, title, threshold=0.7):
|
|
138 |
|
139 |
# Define the Gradio interface
|
140 |
title_input = gr.inputs.Textbox(label="Product Title", placeholder="Enter the product title here...")
|
141 |
-
image_input = gr.inputs.
|
142 |
output = gr.outputs.JSON(label="Top 3 Predictions with Probabilities")
|
143 |
|
144 |
gr.Interface(
|
|
|
6 |
from PIL import Image
|
7 |
from torchvision import transforms
|
8 |
import urllib.request
|
9 |
+
from torchvision import models
|
|
|
10 |
import torch.nn as nn
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
|
12 |
# --- Define the Model ---
|
13 |
class FineGrainedClassifier(nn.Module):
|
|
|
36 |
output = self.classifier(combined_features)
|
37 |
return output
|
38 |
|
|
|
39 |
# --- Data Augmentation Setup ---
|
40 |
transform = transforms.Compose([
|
41 |
transforms.Resize((224, 224)),
|
|
|
46 |
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
|
47 |
])
|
48 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
49 |
# Load the label-to-class mapping from your Hugging Face repository
|
50 |
label_map_url = "https://huggingface.co/Maverick98/EcommerceClassifier/resolve/main/label_to_class.json"
|
51 |
label_to_class = requests.get(label_map_url).json()
|
52 |
|
|
|
53 |
# Load your custom model from Hugging Face
|
54 |
model = FineGrainedClassifier(num_classes=len(label_to_class))
|
|
|
55 |
checkpoint_url = f"https://huggingface.co/Maverick98/EcommerceClassifier/resolve/main/model_checkpoint.pth"
|
56 |
checkpoint = torch.hub.load_state_dict_from_url(checkpoint_url, map_location=torch.device('cpu'))
|
57 |
+
|
58 |
+
# Strip the "module." prefix from the keys in the state_dict if they exist
|
59 |
+
new_state_dict = {}
|
60 |
+
for k, v in checkpoint.items():
|
61 |
+
if k.startswith("module."):
|
62 |
+
new_state_dict[k[7:]] = v # Remove "module." prefix
|
63 |
+
else:
|
64 |
+
new_state_dict[k] = v
|
65 |
+
|
66 |
+
model.load_state_dict(new_state_dict)
|
67 |
|
68 |
# Load the tokenizer from Jina
|
69 |
tokenizer = AutoTokenizer.from_pretrained("jinaai/jina-embeddings-v2-base-en")
|
70 |
|
71 |
+
def load_image(image):
|
|
|
72 |
"""
|
73 |
+
Preprocess the uploaded image.
|
74 |
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
image = transform(image)
|
76 |
image = image.unsqueeze(0) # Add batch dimension
|
77 |
return image
|
78 |
|
79 |
+
def predict(image, title, threshold=0.7):
|
80 |
"""
|
81 |
Predict the top 3 categories for the given image and title.
|
82 |
Includes "Others" if the confidence of the top prediction is below the threshold.
|
83 |
"""
|
84 |
# Preprocess the image
|
85 |
+
image = load_image(image)
|
86 |
|
87 |
# Tokenize the title
|
88 |
title_encoding = tokenizer(title, padding='max_length', max_length=200, truncation=True, return_tensors='pt')
|
|
|
113 |
|
114 |
# Define the Gradio interface
|
115 |
title_input = gr.inputs.Textbox(label="Product Title", placeholder="Enter the product title here...")
|
116 |
+
image_input = gr.inputs.Image(type="pil", label="Upload Image")
|
117 |
output = gr.outputs.JSON(label="Top 3 Predictions with Probabilities")
|
118 |
|
119 |
gr.Interface(
|