app.py

import numpy as np
import matplotlib.pyplot as plt
import cv2
import torch
from segment_anything import sam_model_registry, SamPredictor
from preprocess import show_mask, show_points, show_box
import gradio as gr

sam_checkpoint = {
                    "ViT-base": "weights/sam_vit_b_01ec64.pth",
                    "ViT-large": "weights/sam_vit_l_0b3195.pth",
                    "ViT-huge": "weights/sam_vit_h_4b8939.pth",
                 }
model_type = {
                "ViT-base": "vit_b",
                "ViT-large": "vit_l",
                "ViT-huge": "vit_h",
             }
device = "cuda" if torch.cuda.is_available() else "cpu"


def get_coords(evt: gr.SelectData):
    return f"{evt.index[0]}, {evt.index[1]}"

def inference(image, input_label, model_choice):
    sam = sam_model_registry[model_type[model_choice]](checkpoint=sam_checkpoint[model_choice])
    sam.to(device=device)
    predictor = SamPredictor(sam)
    predictor.set_image(image)
    input_point = np.array([[int(input_label['label'].split(',')[0]), int(input_label['label'].split(',')[1])]])
    input_label = np.array([1])
    masks, scores, logits = predictor.predict(
                                                point_coords=input_point,
                                                point_labels=input_label,
                                                multimask_output=True,
                                             )
    mask = masks[0]
    image2 = image.copy()
    image2[mask, 0] = 255
    return image2


my_app = gr.Blocks()
with my_app:
    gr.Markdown("Segment Anything Testing")
    with gr.Tabs():
        with gr.TabItem("Select your image"):
            with gr.Row():
                with gr.Column():
                    img_source = gr.Image(label="Please select picture and click the part to segment", 
                                          value='./images/truck.jpg', shape=(1024, 1024))
                    coords = gr.Label(label="Image Coordinate")
                    model_choice = gr.Dropdown(['ViT-base', 'ViT-large', 'ViT-huge'], label='Model Backbone')
                    infer = gr.Button(label="Segment")
                with gr.Column():
                    img_output = gr.Image(label="Output Mask", shape=(1024, 1024))
                    
        img_source.select(get_coords, [], coords)
        infer.click(
            inference,
            [
                img_source,
                coords,
                model_choice
            ],
            [
                img_output
            ]
        )
    
my_app.launch(debug=True)