File size: 2,186 Bytes
5287bf1 bfd8285 5287bf1 358b6e7 bfd8285 5287bf1 6d7226f bfd8285 5287bf1 bfd8285 6d7226f bfd8285 5287bf1 bfd8285 5287bf1 6d7226f f4b23ea 5287bf1 bfd8285 5287bf1 6d7226f e36d373 5287bf1 bfd8285 |
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 |
#!/usr/bin/env python
import torch
import numpy as np
import gradio as gr
from faiss import read_index
from PIL import Image, ImageOps
from datasets import load_dataset
import torchvision.transforms as T
from model import DINO
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
## Define Model and Dataset
dataset = load_dataset("ethz/food101")
model = DINO(batch_size_per_device=32, num_classes=1000).to(device)
model.load_state_dict(torch.load("./bin/model.ckpt", map_location=device)["state_dict"])
def augment(img: np.ndarray) -> torch.Tensor:
"""
Helper Function to augment the image before we generate embeddings
Args:
img (np.ndarray): Input Image
Returns:
torch.Tensor
"""
img = Image.fromarray(img)
if img.mode == "L":
# Convert grayscale image to RGB by duplicating the single channel three times
img = ImageOps.colorize(img, black="black", white="white")
transforms = T.Compose(
[T.ToTensor(), T.Resize(244), T.CenterCrop(224), T.Normalize([0.5], [0.5])]
)
return transforms(img).unsqueeze(0)
def search_index(input_image: np.ndarray, k: int = 1) -> list:
"""
Retrieve the Top k images from the given input image
Args:
input_image (np.ndarray): Input Image
k (int): number of images to fetch
Returns:
list: List of top k images retrieved using the embeddings
generated from the input image
"""
images = []
with torch.no_grad():
embedding = model(augment(input_image).to(device))
index = read_index("./bin/dino.index")
_, results = index.search(np.array(embedding[0].reshape(1, -1)), k)
indices = results[0]
for _, index in enumerate(indices[:k]):
retrieved_img = dataset["train"][int(index)]["image"]
images.append(retrieved_img)
return images
app = gr.Interface(
search_index,
inputs=[
gr.Image(label="Input Image"),
gr.Slider(minimum=1, maximum=10, value=4, step=1, label="Top K"),
],
outputs=[
gr.Gallery(label="Retrieved Images"),
],
)
if __name__ == "__main__":
app.launch()
|