File size: 5,849 Bytes
dfbfc84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f634c61
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
import gradio as gr
import torch
import torch.nn as nn
import torch.nn.functional as F
from transformers import ViTModel, AutoModel, AutoTokenizer
from torchvision import transforms
from datasets import load_dataset
from PIL import Image

class MultiModalEngine(nn.Module):
    def __init__(self):
        super().__init__()
        self.image_model = ViTModel.from_pretrained("google/vit-base-patch16-224")
        self.text_model = AutoModel.from_pretrained("sentence-transformers/all-mpnet-base-v2")
        self.image_projection = nn.Linear(768, 256)
        self.text_projection = nn.Linear(768, 256)
        self.logit_scale = nn.Parameter(torch.ones([]) * 2.659)

    def encode_text(self, input_ids, attention_mask):
        text_out = self.text_model(input_ids=input_ids, attention_mask=attention_mask)
        text_embeds = self.text_projection(self.mean_pooling(text_out, attention_mask))
        return F.normalize(text_embeds, dim=1)

    def encode_image(self, images):
        vision_out = self.image_model(pixel_values=images)
        image_embeds = self.image_projection(vision_out.last_hidden_state[:, 0, :])
        return F.normalize(image_embeds, dim=1)

    def mean_pooling(self, model_output, attention_mask):
        token_embeddings = model_output.last_hidden_state
        mask = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
        return torch.sum(token_embeddings * mask, 1) / torch.clamp(mask.sum(1), min=1e-9)


print("⏳ Loading resources...")
device = "cpu"

# Load Model
model = MultiModalEngine()
model.load_state_dict(torch.load("flickr8k_best_model_r1_27.pth", map_location=device))
model.eval()

# Load Index
image_embeddings = torch.load("flickr8k_best_index.pt", map_location=device)

# Load Tokenizer & Transforms
tokenizer = AutoTokenizer.from_pretrained("sentence-transformers/all-mpnet-base-v2")
val_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])
])
# Load Dataset (Standard mode to fetch result images)
print("Downloading dataset (this may take a minute)...")
dataset = load_dataset("tsystems/flickr8k", split="train")

print("Server Ready!")


def search_text(query):
    inputs = tokenizer(query, return_tensors="pt", padding=True, truncation=True)
    with torch.no_grad():
        text_emb = model.encode_text(inputs['input_ids'], inputs['attention_mask'])
    
    scores = text_emb @ image_embeddings.T
    scores = scores.squeeze()
    values, indices = torch.topk(scores, 3)
    
    return [dataset[int(idx)]['image'] for idx in indices]

def search_image(query_img):
    if query_img is None: return []
    # Ensure it's a PIL Image
    if not isinstance(query_img, Image.Image):
        query_img = Image.fromarray(query_img)
        
    img_tensor = val_transform(query_img).unsqueeze(0)
    with torch.no_grad():
        img_emb = model.encode_image(img_tensor)
    
    scores = img_emb @ image_embeddings.T
    scores = scores.squeeze()
    values, indices = torch.topk(scores, 3)
    
    return [dataset[int(idx)]['image'] for idx in indices]


with gr.Blocks(title="CLIP Sytle MultiModal Search", theme=gr.themes.Soft()) as demo:
    gr.Markdown("# 🔍CLIP Sytle MultiModal")
    gr.Markdown("Search for images using **Text** OR using another **Image**.")
    
    with gr.Tabs():
        # --- TAB 1: TEXT SEARCH ---
        with gr.TabItem("Search by Text"):
            with gr.Row():
                txt_input = gr.Textbox(label="Type your query", placeholder="e.g. A dog running...")
                txt_btn = gr.Button("Search", variant="primary")
            
            txt_gallery = gr.Gallery(label="Top Matches", columns=3, height=300)
            
            # CLICKABLE TEXT EXAMPLES
            gr.Examples(
                examples=[
                    ["A dog running on grass"],
                    ["Children playing in the water"],
                    ["A girl in a pink dress"],
                    ["A man climbing a rock"]
                ],
                inputs=txt_input,       # Clicking populates this box
                outputs=txt_gallery,    # Result appears here
                fn=search_text,         # Function to run
                run_on_click=True,      # Run immediately when clicked!
                label="Try these examples:"
            )
            
            txt_btn.click(search_text, inputs=txt_input, outputs=txt_gallery)

        # --- TAB 2: IMAGE SEARCH ---
        with gr.TabItem("Search by Image"):
            # Define components first (but don't draw them yet)
            # We set render=False so we can place them visually later
            img_input = gr.Image(type="pil", label="Upload Source Image", sources=['upload', 'clipboard'], render=False)
            img_gallery = gr.Gallery(label="Similar Images", columns=3, height=300, render=False)
            
            # Draw Examples FIRST (So they appear at the very top)
            gr.Examples(
                examples=[
                    ["examples/dog.jpg"], 
                    ["examples/beach.jpg"]
                ],
                inputs=img_input,
                outputs=img_gallery,
                fn=search_image,
                run_on_click=True,
                label="Click an image to test:"
            )
            
            # Draw Input and Button (Visually below examples)
            with gr.Row():
                img_input.render() #
                img_btn = gr.Button("Find Similar", variant="primary")
            
            # Draw Gallery (Visually at the bottom)
            img_gallery.render()
            
            # Connect the Button
            img_btn.click(search_image, inputs=img_input, outputs=img_gallery)

if __name__ == "__main__":
    demo.launch()