app.py

import gradio as gr
import os, requests
import numpy as np
import torch
import cv2
from cell_segmentation.inference.inference_cellvit_experiment_pannuke import InferenceCellViTParser,InferenceCellViT
from cell_segmentation.inference.inference_cellvit_experiment_monuseg import InferenceCellViTMoNuSegParser,MoNuSegInference


## local |  remote
RUN_MODE = "remote"
if RUN_MODE != "local":
    os.system("wget https://huggingface.co/xiazhi/LKCell/resolve/main/model_best.pth")
    ## examples
    os.system("wget https://huggingface.co/xiazhi/LKCell/resolve/main/1.png")
    os.system("wget https://huggingface.co/xiazhi/LKCell/resolve/main/2.png")
    os.system("wget https://huggingface.co/xiazhi/LKCell/resolve/main/3.png")
    os.system("wget https://huggingface.co/xiazhi/LKCell/resolve/main/4.png")

## step 1: set up model

device = "cpu"

## pannuke set
pannuke_parser = InferenceCellViTParser()
pannuke_configurations = pannuke_parser.parse_arguments()
pannuke_inf = InferenceCellViT(
        run_dir=pannuke_configurations["run_dir"],
        checkpoint_name=pannuke_configurations["checkpoint_name"],
        gpu=pannuke_configurations["gpu"],
        magnification=pannuke_configurations["magnification"],
    )

pannuke_checkpoint = torch.load(
    pannuke_inf.run_dir  / pannuke_inf.checkpoint_name, map_location="cpu"
)
pannuke_model = pannuke_inf.get_model(model_type=pannuke_checkpoint["arch"])
pannuke_model.load_state_dict(pannuke_checkpoint["model_state_dict"])
# # put model in eval mode
pannuke_model.to(device)
pannuke_model.eval()


## monuseg set
monuseg_parser = InferenceCellViTMoNuSegParser()
monuseg_configurations = monuseg_parser.parse_arguments()
monuseg_inf = MoNuSegInference(
        model_path=monuseg_configurations["model"],
        dataset_path=monuseg_configurations["dataset"],
        outdir=monuseg_configurations["outdir"],
        gpu=monuseg_configurations["gpu"],
        patching=monuseg_configurations["patching"],
        magnification=monuseg_configurations["magnification"],
        overlap=monuseg_configurations["overlap"],
    )


def click_process(image_input , type_dataset):
    if type_dataset == "pannuke":
        if image_input.shape[0] > 512 and image_input.shape[1] > 512:
            image_input = cv2.resize(image_input, (512,512))
        pannuke_inf.run_single_image_inference(pannuke_model,image_input)
    else:
        if image_input.shape[0] > 512 and image_input.shape[1] > 512:
            image_input = cv2.resize(image_input, (512,512))
        monuseg_inf.run_single_image_inference(monuseg_inf.model, image_input)

    image_output = cv2.imread("raw_pred.png")
    image_output = cv2.cvtColor(image_output, cv2.COLOR_BGR2RGB)     
    image_output2 = cv2.imread("pred_img.png")
    image_output2 = cv2.cvtColor(image_output2, cv2.COLOR_BGR2RGB)   
    return image_output,image_output2


demo = gr.Blocks(title="LkCell")
with demo:
    gr.Markdown(value="""
                    **Gradio demo for LKCell: Efficient Cell Nuclei Instance Segmentation with Large Convolution Kernels**. Check our [Github Repo](https://github.com/hustvl/LKCell) 😛.
                    """)
    with gr.Row():
        with gr.Column():
            with gr.Row():
                Image_input = gr.Image(type="numpy", label="Input", interactive=True,height=480)
            with gr.Row():
                Type_dataset = gr.Radio(choices=["pannuke", "monuseg"], label=" input image's dataset type",value="pannuke")
                                            
        with gr.Column():
            image_output = gr.Image(type="numpy", label="image prediction",height=480,width=480)
            image_output2 = gr.Image(type="numpy", label="all predictions",height=480)
            
    with gr.Row():
        Button_run = gr.Button("🚀 Submit (发送) ")
        clear_button = gr.ClearButton(components=[Image_input,Type_dataset,image_output,image_output2],value="🧹 Clear (清除)")

    Button_run.click(fn=click_process, inputs=[Image_input,  Type_dataset ], outputs=[image_output,image_output2])
    
    ## guiline
    gr.Markdown(value="""    
                    🔔**Guideline**
                    1. Upload your image or select one from the examples.
                    2. Set up the arguments: "Type_dataset" to enjoy two dataset type's inference
                    3. Due to the limit of CPU , we resize the input image whose size is larger than (512,512) to (512,512)
                    4. Run the Submit button to get the output.
                    """)
    # if RUN_MODE != "local":
    gr.Examples(examples=[
                ['1.png', "pannuke"],
                ['2.png', "pannuke"],
                ['3.png', "monuseg"],
                ['4.png', "monuseg"],
                ], 
                inputs=[Image_input, Type_dataset], outputs=[image_output,image_output2], label="Examples")
    gr.HTML(value="""
                <p style="text-align:center; color:orange"> <a href='https://github.com/hustvl/LKCell' target='_blank'>Github Repo</a></p>
                    """)
    gr.Markdown(value="""    
                    Template is adapted from [Here](https://huggingface.co/spaces/menghanxia/disco)
                    """)

if RUN_MODE == "local":
    demo.launch(server_name='127.0.0.1',server_port=8003)
else:
    demo.launch()