preprocess/humanparsing/run_parsing.py

import pdb
from pathlib import Path
import sys
import os
import onnxruntime as ort
PROJECT_ROOT = Path(__file__).absolute().parents[0].absolute()
sys.path.insert(0, str(PROJECT_ROOT))
from parsing_api import onnx_inference
import torch
import requests



# class Parsing:
#     def __init__(self, gpu_id: int):
#         self.gpu_id = gpu_id
#         torch.cuda.set_device(gpu_id)
#         session_options = ort.SessionOptions()
#         session_options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_ALL
#         session_options.execution_mode = ort.ExecutionMode.ORT_SEQUENTIAL
#         session_options.add_session_config_entry('gpu_id', str(gpu_id))
#         self.session = ort.InferenceSession(os.path.join(Path(__file__).absolute().parents[2].absolute(), 'ckpt/humanparsing/parsing_atr.onnx'),
#                                             sess_options=session_options, providers=['CPUExecutionProvider'])
#         self.lip_session = ort.InferenceSession(os.path.join(Path(__file__).absolute().parents[2].absolute(), 'ckpt/humanparsing/parsing_lip.onnx'),
#                                                 sess_options=session_options, providers=['CPUExecutionProvider'])
        

#     def __call__(self, input_image):
#         # torch.cuda.set_device(self.gpu_id)
#         parsed_image, face_mask = onnx_inference(self.session, self.lip_session, input_image)
#         return parsed_image, face_mask



class Parsing:
    def __init__(self, gpu_id: int):
        self.gpu_id = gpu_id
        torch.cuda.set_device(gpu_id)
        
        # Define the URLs for the models on Hugging Face
        atr_url = 'https://huggingface.co/basso4/humanparsing/resolve/main/parsing_atr.onnx'
        lip_url = 'https://huggingface.co/basso4/humanparsing/resolve/main/parsing_lip.onnx'
        
        # Define local paths for storing the models
        model_dir = Path.home() / '.cache/humanparsing_models'
        model_dir.mkdir(parents=True, exist_ok=True)
        atr_path = model_dir / 'parsing_atr.onnx'
        lip_path = model_dir / 'parsing_lip.onnx'
        
        # Download models if they don't already exist locally
        self.download_model(atr_url, atr_path)
        self.download_model(lip_url, lip_path)

        # Configure ONNX Runtime session options
        session_options = ort.SessionOptions()
        session_options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_ALL
        session_options.execution_mode = ort.ExecutionMode.ORT_SEQUENTIAL
        session_options.add_session_config_entry('gpu_id', str(gpu_id))
        
        # Initialize ONNX Inference Sessions
        self.session = ort.InferenceSession(str(atr_path), sess_options=session_options, providers=['CPUExecutionProvider'])
        self.lip_session = ort.InferenceSession(str(lip_path), sess_options=session_options, providers=['CPUExecutionProvider'])

    def download_model(self, url, path):
        # Download the model if it doesn't exist at the specified path
        if not path.exists():
            print(f"Downloading model from {url}...")
            response = requests.get(url, stream=True)
            response.raise_for_status()
            with open(path, 'wb') as file:
                for chunk in response.iter_content(chunk_size=8192):
                    file.write(chunk)
            print(f"Model downloaded and saved to {path}")

    def __call__(self, input_image):
        # Perform inference using both ONNX models and return the results
        parsed_image, face_mask = onnx_inference(self.session, self.lip_session, input_image)
        return parsed_image, face_mask