|
import torch
|
|
from torchvision import models, transforms
|
|
from PIL import Image
|
|
import numpy as np
|
|
|
|
|
|
def segment_person(image_path):
|
|
|
|
model = models.segmentation.deeplabv3_resnet101(pretrained=True).eval()
|
|
|
|
|
|
input_image = Image.open(image_path).convert("RGB")
|
|
preprocess = transforms.Compose(
|
|
[
|
|
transforms.ToTensor(),
|
|
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
|
|
]
|
|
)
|
|
input_tensor = preprocess(input_image).unsqueeze(0)
|
|
|
|
with torch.no_grad():
|
|
output = model(input_tensor)["out"][0]
|
|
mask = output.argmax(0).byte().numpy()
|
|
|
|
|
|
segmented_image = np.array(input_image)
|
|
segmented_image = np.dstack([segmented_image, mask * 255])
|
|
return Image.fromarray(segmented_image)
|
|
|