app.py

import torch
import numpy as np
import gradio as gr
import torch.nn.functional as F
from PIL import Image
from fastsam import FastSAM, FastSAMPrompt

device = 'cpu'
if torch.cuda.is_available():
    device = 'cuda'
elif hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
    device = "mps"

model = FastSAM('./weights/FastSAM-x.pt')
model.to(device)

def inference(image, conf_thres, iou_thres, text):
    pred = model(image, device=device, retina_masks=True, imgsz=1024, conf=conf_thres, iou=iou_thres)
    prompt_process = FastSAMPrompt(image, pred, device="cpu")
    ann = prompt_process.everything_prompt()
    prompt_process.plot(annotations=ann, output_path="./output_sam.jpg", withContours=False, better_quality=False)
    output_sam = Image.open('./output_sam.jpg')
    output_sam = np.array(output_sam)
    output_text = None
    if text:
        ann = prompt_process.text_prompt(text=text)
        prompt_process.plot(annotations=ann, output_path="./output_text.jpg", withContours=False, better_quality=False)
        output_text = Image.open('./output_text.jpg')
        output_text = np.array(output_text)
    return output_sam, output_text

title = "FAST-SAM Segment Anything"
description = "A simple Gradio interface to infer on FAST-SAM model"
examples = [["image_1.jpg", 0.25, 0.45, 'A black tire'],
            ["image_2.jpg", 0.25, 0.45, 'Shades of blue'],
            ["image_3.jpg", 0.25, 0.45, 'A spiral staircase'],
            ["image_4.jpg", 0.25, 0.45, 'A clock and a plane'],
            ["image_5.jpg", 0.25, 0.45, 'Clouds in the sky'],
            ["image_6.jpg", 0.25, 0.45, 'Front wheel'],
            ["image_7.jpg", 0.25, 0.45, 'A white chair'],
            ["image_8.jpg", 0.25, 0.45, 'The grassy field'],
            ["image_9.jpg", 0.25, 0.45, 'Rock formation'],
            ["image_10.jpg", 0.25, 0.45, 'A rope railing']]

demo = gr.Interface(inference, 
                    inputs = [gr.Image(width=320, height=320, label="Input Image"), 
                              gr.Slider(0, 1, 0.25, label="Confidence Threshold"),
                              gr.Slider(0, 1, 0.45, label="IoU Thresold"),
                              gr.Textbox(label="Enter text promp", type="text"),],
                    outputs= [gr.Image(width=320, height=320, label="Output SAM"),
                              gr.Image(width=320, height=320, label="Output text prompt")],
                    title=title,
                    description=description,
                    examples=examples)

demo.launch()