main.py

from fastapi import FastAPI, File, UploadFile
from fastapi.responses import StreamingResponse
import numpy as np
import torch
import torch.nn.functional as F
from torchvision.transforms.functional import normalize
from briarmbg import BriaRMBG
from PIL import Image
import io

# Initialize the model
net = BriaRMBG.from_pretrained("briaai/RMBG-1.4")
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
net.to(device)
net.eval()

# Initialize FastAPI app
app = FastAPI()

def resize_image(image):
    image = image.convert('RGB')
    model_input_size = (1024, 1024)
    image = image.resize(model_input_size, Image.BILINEAR)
    return image

def process_image(image):
    # prepare input
    orig_image = Image.open(image.file).convert("RGB")
    w, h = orig_im_size = orig_image.size
    image = resize_image(orig_image)
    im_np = np.array(image)
    im_tensor = torch.tensor(im_np, dtype=torch.float32).permute(2, 0, 1)
    im_tensor = torch.unsqueeze(im_tensor, 0)
    im_tensor = torch.divide(im_tensor, 255.0)
    im_tensor = normalize(im_tensor, [0.5, 0.5, 0.5], [1.0, 1.0, 1.0])
    if torch.cuda.is_available():
        im_tensor = im_tensor.cuda()

    # inference
    result = net(im_tensor)
    # post process
    result = torch.squeeze(F.interpolate(result[0][0], size=(h, w), mode='bilinear'), 0)
    ma = torch.max(result)
    mi = torch.min(result)
    result = (result - mi) / (ma - mi)
    # image to pil
    result_array = (result * 255).cpu().data.numpy().astype(np.uint8)
    pil_mask = Image.fromarray(np.squeeze(result_array))
    # add the mask on the original image as alpha channel
    new_im = orig_image.copy()
    new_im.putalpha(pil_mask)
    return new_im

@app.post("/process-image/")
async def process(file: UploadFile = File(...)):
    processed_image = process_image(file)
    # Save the processed image to a bytes buffer
    buf = io.BytesIO()
    processed_image.save(buf, format="PNG")
    buf.seek(0)
    return StreamingResponse(buf, media_type="image/png")