Help find dependencies, python 3.9 problem with conversion Converting ONNX -> TensorFlow

Abierto
  • ¿Algo no está bien?
#2,004 0 comentarios 0 reacciones 0 asignados Ver en GitHub

Nadie ha tomado este issue todavía.

Evaluación

Dificultad
4/5
Tiempo estimado
3-5 días
Aptitud para principiantes
25/100
Tipo de issue
Error
Claridad
Necesita aclaración
Estado de actividad
Estancado
Stack tecnológico
python, pytorch, tensorflow
Área
machine-learning, tooling

Línea de trabajo

Comienza con el script de Python proporcionado y reproduce el paso convert_onnx_to_tf, anotando las versiones de Python y de las dependencias utilizadas. El trabajo estará terminado cuando se haya identificado un conjunto de dependencias compatible y se haya documentado como un archivo requirements funcional o una especificación de entorno equivalente.

Escrito por el modelo de indexación a partir del texto del issue.

Descripción

Please help me resolve dependency issues.

I'm using Python 3.9 and encountering problems during the conversion step Converting ONNX -> TensorFlow. No matter what I try, I keep running into conflicts or other errors. I cannot find a proper set of dependencies or at least a working requirements.txt.

Some sources say Python 3.9 is required, others say 3.8, and some recommend 3.10. Honestly, I’m exhausted trying to figure this out.

import os
import subprocess
import torch
import onnx
import onnxslim
import sys
import types
from pathlib import Path
import urllib.request
import zipfile
import shutil

def download_yolov5_repo(target_dir):
    zip_url = "https://github.com/ultralytics/yolov5/archive/refs/heads/master.zip"
    zip_path = os.path.join(target_dir, "yolov5.zip")
    print("📥 Downloading YOLOv5 repository...")
    urllib.request.urlretrieve(zip_url, zip_path)

    with zipfile.ZipFile(zip_path, 'r') as zip_ref:
        zip_ref.extractall(target_dir)
    os.remove(zip_path)

    extracted_path = os.path.join(target_dir, "yolov5-master")
    yolov5_path = os.path.join(target_dir, "yolov5")
    if os.path.exists(yolov5_path):
        shutil.rmtree(yolov5_path)
    shutil.move(extracted_path, yolov5_path)
    print("✅ YOLOv5 repository downloaded and prepared.")

def convert_pt_to_onnx(pt_path, onnx_path):
    yolov5_dir = os.path.join(Path(pt_path).parent, "yolov5")
    yolov5_model_path = os.path.join(yolov5_dir, "models", "experimental.py")

    if not os.path.exists(yolov5_model_path):
        print("⚠️ YOLOv5 repo not found, downloading it...")
        download_yolov5_repo(Path(pt_path).parent)

    if yolov5_dir not in sys.path:
        sys.path.insert(0, yolov5_dir)

    try:
        from models.experimental import attempt_load
    except ImportError as e:
        print("❌ ERROR: Could not import YOLOv5 modules. Ensure YOLOv5 repo is in the correct path.")
        raise e

    model = attempt_load(pt_path, device=torch.device('cpu'))
    model.eval()

    dummy_input = torch.randn(1, 3, 640, 640)

    torch.onnx.export(
        model,
        dummy_input,
        onnx_path,
        export_params=True,
        opset_version=12,
        do_constant_folding=True,
        input_names=['images'],
        output_names=['output'],
        dynamic_axes={'images': {0: 'batch_size'}, 'output': {0: 'batch_size'}}
    )

    onnxslim.slim(onnx_path, onnx_path)

def convert_onnx_to_tf(onnx_path, tf_path):
    subprocess.run(['onnx-tf', 'convert', '-i', onnx_path, '-o', tf_path], check=True)

def convert_tf_to_tfjs(tf_path, tfjs_path):
    try:
        print("⚙️  Importing TensorFlow.js converter inside Python...")
        sys.modules['tensorflow_decision_forests'] = types.SimpleNamespace()
        sys.modules['tensorflow_decision_forests.keras'] = types.SimpleNamespace()
        sys.modules['tensorflow_decision_forests.keras.core'] = types.SimpleNamespace()
        sys.modules['tensorflow_decision_forests.keras.core_inference'] = types.SimpleNamespace()
        sys.modules['tensorflow_decision_forests.tensorflow'] = types.SimpleNamespace()
        sys.modules['tensorflow_decision_forests.tensorflow.ops'] = types.SimpleNamespace()
        sys.modules['tensorflow_decision_forests.tensorflow.ops.inference'] = types.SimpleNamespace()
        sys.modules['tensorflow_decision_forests.tensorflow.ops.inference.api'] = types.SimpleNamespace()
        sys.modules['tensorflow_decision_forests.tensorflow.ops.inference.op'] = types.SimpleNamespace()
        sys.modules['tensorflow_decision_forests.tensorflow.ops.inference.op_dynamic'] = types.SimpleNamespace()

        import tensorflowjs as tfjs

        tfjs.converters.convert_tf_saved_model(
            saved_model_dir=tf_path,
            output_dir=tfjs_path,
            skip_op_check=True,
            strip_debug_ops=True
        )
        print(f"✅ TensorFlow.js conversion complete! Files saved to: {tfjs_path}")
    except Exception as e:
        print("\n❌ ERROR: TensorFlow.js conversion failed unexpectedly.")
        raise e

def main():
    folder_path = input("Enter the folder path to search for PyTorch (.pt) models: ")
    pt_files = [f for f in os.listdir(folder_path) if f.endswith(".pt")]

    if not pt_files:
        print("No PyTorch (.pt) models found.")
        return

    print("\nFound PyTorch models:")
    for i, f in enumerate(pt_files, start=1):
        print(f"{i}. {f}")

    idx = int(input("Enter the number of the model you want to convert: ")) - 1
    pt_path = os.path.join(folder_path, pt_files[idx])
    model_name = os.path.splitext(pt_files[idx])[0]

    onnx_path = os.path.join(folder_path, f"{model_name}.onnx")
    tf_path = os.path.join(folder_path, "model_tf")
    tfjs_path = os.path.join(folder_path, f"{model_name}WebModel")

    print("\nConverting PyTorch -> ONNX using YOLOv5...")
    convert_pt_to_onnx(pt_path, onnx_path)

    print("\nConverting ONNX -> TensorFlow...")
    convert_onnx_to_tf(onnx_path, tf_path)

    print("\nConverting TensorFlow -> TensorFlow.js...")
    convert_tf_to_tfjs(tf_path, tfjs_path)

if __name__ == '__main__':
    main()
Lenguaje dominante
Jupyter Notebook
Estrellas
4.4k
Forks
1.1k
Sin PR fusionados en 30 d

Preparar el entorno

Primeros pasos

  1. Lee el issue completo y luego la guía de contribución del proyecto.
  2. Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
  3. Haz un fork del repositorio y trabaja en una rama.
  4. Abre un pull request que haga referencia al número del issue.

Más de tensorflow/probability

Todos los issues de tensorflow/probability

Issues similares

Más issues de Machine Learning

Recibe los nuevos issues en tu correo

Un resumen breve de issues de GitHub para principiantes.