scripts/rendering/main.py

import glob
import json
import multiprocessing
import os
import platform
import random
import subprocess
import tempfile
import time
import zipfile
from functools import partial
from typing import Any, Callable, Dict, List, Literal, Optional, Union

import fire
import fsspec
import GPUtil
import pandas as pd
from loguru import logger

from objaverse_xl.github import download_github_objects
from objaverse_xl.utils import get_uid_from_str


def log_processed_object(csv_filename: str, *args) -> None:
    """Log when an object is done being used.

    Args:
        csv_filename (str): Name of the CSV file to save the logs to.
        *args: Arguments to save to the CSV file.

    Returns:
        None
    """
    args = ",".join([str(arg) for arg in args])
    # log that this object was rendered successfully
    # saving locally to avoid excessive writes to the cloud
    dirname = os.path.expanduser(f"~/.objaverse/github/logs/")
    os.makedirs(dirname, exist_ok=True)
    with open(os.path.join(dirname, csv_filename), "a", encoding="utf-8") as f:
        f.write(f"{time.time()},{args}\n")


def zipdir(path: str, ziph: zipfile.ZipFile) -> None:
    """Zip up a directory with an arcname structure.

    Args:
        path (str): Path to the directory to zip.
        ziph (zipfile.ZipFile): ZipFile handler object to write to.

    Returns:
        None
    """
    # ziph is zipfile handle
    for root, dirs, files in os.walk(path):
        for file in files:
            # this ensures the structure inside the zip starts at folder/
            arcname = os.path.join(os.path.basename(root), file)
            ziph.write(os.path.join(root, file), arcname=arcname)


def handle_found_object(
    local_path: str,
    file_identifier: str,
    sha256: str,
    metadata: Dict[str, Any],
    num_renders: int,
    render_dir: str,
    only_northern_hemisphere: bool,
    gpu_devices: Union[int, List[int]],
    render_timeout: int,
    successful_log_file: Optional[str] = "handle-found-object-successful.csv",
    failed_log_file: Optional[str] = "handle-found-object-failed.csv",
) -> bool:
    """Called when an object is successfully found and downloaded.

    Here, the object has the same sha256 as the one that was downloaded with
    Objaverse-XL. If None, the object will be downloaded, but nothing will be done with
    it.

    Args:
        local_path (str): Local path to the downloaded 3D object.
        file_identifier (str): GitHub URL of the 3D object.
        sha256 (str): SHA256 of the contents of the 3D object.
        metadata (Dict[str, Any]): Metadata about the 3D object, including keys for
            `github_organization` and `github_repo`.
        num_renders (int): Number of renders to save of the object.
        render_dir (str): Directory where the objects will be rendered.
        only_northern_hemisphere (bool): Only render the northern hemisphere of the
            object.
        gpu_devices (Union[int, List[int]]): GPU device(s) to use for rendering. If
            an int, the GPU device will be randomly selected from 0 to gpu_devices - 1.
            If a list, the GPU device will be randomly selected from the list.
            If 0, the CPU will be used for rendering.
        render_timeout (int): Number of seconds to wait for the rendering job to
            complete.
        successful_log_file (str): Name of the log file to save successful renders to.
        failed_log_file (str): Name of the log file to save failed renders to.

    Returns: True if the object was rendered successfully, False otherwise.
    """
    save_uid = get_uid_from_str(file_identifier)
    args = f"--object_path '{local_path}' --num_renders {num_renders}"

    # get the GPU to use for rendering
    using_gpu: bool = True
    gpu_i = 0
    if isinstance(gpu_devices, int) and gpu_devices > 0:
        num_gpus = gpu_devices
        gpu_i = random.randint(0, num_gpus - 1)
    elif isinstance(gpu_devices, list):
        gpu_i = random.choice(gpu_devices)
    elif isinstance(gpu_devices, int) and gpu_devices == 0:
        using_gpu = False
    else:
        raise ValueError(
            f"gpu_devices must be an int > 0, 0, or a list of ints. Got {gpu_devices}."
        )

    with tempfile.TemporaryDirectory() as temp_dir:
        # get the target directory for the rendering job
        target_directory = os.path.join(temp_dir, save_uid)
        os.makedirs(target_directory, exist_ok=True)
        args += f" --output_dir {target_directory}"

        # check for Linux / Ubuntu or MacOS
        if platform.system() == "Linux" and using_gpu:
            args += " --engine BLENDER_EEVEE"
        elif platform.system() == "Darwin" or (
            platform.system() == "Linux" and not using_gpu
        ):
            # As far as I know, MacOS does not support BLENER_EEVEE, which uses GPU
            # rendering. Generally, I'd only recommend using MacOS for debugging and
            # small rendering jobs, since CYCLES is much slower than BLENDER_EEVEE.
            args += " --engine CYCLES"
        else:
            raise NotImplementedError(f"Platform {platform.system()} is not supported.")

        # check if we should only render the northern hemisphere
        if only_northern_hemisphere:
            args += " --only_northern_hemisphere"

        # get the command to run
        command = f"blender-3.2.2-linux-x64/blender --background --python blender_script.py -- {args}"
        if using_gpu:
            command = f"export DISPLAY=:0.{gpu_i} && {command}"

        # render the object (put in dev null)
        subprocess.run(
            ["bash", "-c", command],
            timeout=render_timeout,
            check=False,
            stdout=subprocess.DEVNULL,
            stderr=subprocess.DEVNULL,
        )

        # check that the renders were saved successfully
        png_files = glob.glob(os.path.join(target_directory, "*.png"))
        metadata_files = glob.glob(os.path.join(target_directory, "*.json"))
        npy_files = glob.glob(os.path.join(target_directory, "*.npy"))
        if (
            (len(png_files) != num_renders)
            or (len(npy_files) != num_renders)
            or (len(metadata_files) != 1)
        ):
            logger.error(
                f"Found object {file_identifier} was not rendered successfully!"
            )
            if failed_log_file is not None:
                log_processed_object(
                    failed_log_file,
                    file_identifier,
                    sha256,
                )
            return False

        # update the metadata
        metadata_path = os.path.join(target_directory, "metadata.json")
        with open(metadata_path, "r", encoding="utf-8") as f:
            metadata_file = json.load(f)
        metadata_file["sha256"] = sha256
        metadata_file["file_identifier"] = file_identifier
        metadata_file["save_uid"] = save_uid
        metadata_file["metadata"] = metadata
        with open(metadata_path, "w", encoding="utf-8") as f:
            json.dump(metadata, f, indent=2, sort_keys=True)

        # Make a zip of the target_directory.
        # Keeps the {save_uid} directory structure when unzipped
        with zipfile.ZipFile(
            f"{target_directory}.zip", "w", zipfile.ZIP_DEFLATED
        ) as ziph:
            zipdir(target_directory, ziph)

        # move the zip to the render_dir
        fs, path = fsspec.core.url_to_fs(render_dir)

        # move the zip to the render_dir
        fs.makedirs(os.path.join(path, "github", "renders"), exist_ok=True)
        fs.put(
            os.path.join(f"{target_directory}.zip"),
            os.path.join(path, "github", "renders", f"{save_uid}.zip"),
        )

        # log that this object was rendered successfully
        if successful_log_file is not None:
            log_processed_object(successful_log_file, file_identifier, sha256)

        return True


def handle_new_object(
    local_path: str,
    file_identifier: str,
    sha256: str,
    metadata: Dict[str, Any],
    log_file: str = "handle-new-object.csv",
) -> None:
    """Called when a new object is found.

    Here, the object is not used in Objaverse-XL, but is still downloaded with the
    repository. The object may have not been used because it does not successfully
    import into Blender. If None, the object will be downloaded, but nothing will be
    done with it.

    Args:
        local_path (str): Local path to the downloaded 3D object.
        file_identifier (str): GitHub URL of the 3D object.
        sha256 (str): SHA256 of the contents of the 3D object.
        metadata (Dict[str, Any]): Metadata about the 3D object, including the GitHub
            organization and repo names.
        log_file (str): Name of the log file to save the handle_new_object logs to.

    Returns:
        None
    """
    # log the new object
    log_processed_object(log_file, file_identifier, sha256)


def handle_modified_object(
    local_path: str,
    file_identifier: str,
    new_sha256: str,
    old_sha256: str,
    metadata: Dict[str, Any],
    num_renders: int,
    render_dir: str,
    only_northern_hemisphere: bool,
    gpu_devices: Union[int, List[int]],
    render_timeout: int,
) -> None:
    """Called when a modified object is found and downloaded.

    Here, the object is successfully downloaded, but it has a different sha256 than the
    one that was downloaded with Objaverse-XL. This is not expected to happen very
    often, because the same commit hash is used for each repo. If None, the object will
    be downloaded, but nothing will be done with it.

    Args:
        local_path (str): Local path to the downloaded 3D object.
        file_identifier (str): GitHub URL of the 3D object.
        new_sha256 (str): SHA256 of the contents of the newly downloaded 3D object.
        old_sha256 (str): Expected SHA256 of the contents of the 3D object as it was
            when it was downloaded with Objaverse-XL.
        metadata (Dict[str, Any]): Metadata about the 3D object, including the GitHub
            organization and repo names.
        num_renders (int): Number of renders to save of the object.
        render_dir (str): Directory where the objects will be rendered.
        only_northern_hemisphere (bool): Only render the northern hemisphere of the
            object.
        gpu_devices (Union[int, List[int]]): GPU device(s) to use for rendering. If
            an int, the GPU device will be randomly selected from 0 to gpu_devices - 1.
            If a list, the GPU device will be randomly selected from the list.
            If 0, the CPU will be used for rendering.
        render_timeout (int): Number of seconds to wait for the rendering job to
            complete.

    Returns:
        None
    """
    success = handle_found_object(
        local_path=local_path,
        file_identifier=file_identifier,
        sha256=new_sha256,
        metadata=metadata,
        num_renders=num_renders,
        render_dir=render_dir,
        only_northern_hemisphere=only_northern_hemisphere,
        gpu_devices=gpu_devices,
        render_timeout=render_timeout,
        successful_log_file=None,
        failed_log_file=None,
    )

    if success:
        log_processed_object(
            "handle-modified-object-successful.csv",
            file_identifier,
            old_sha256,
            new_sha256,
        )
    else:
        log_processed_object(
            "handle-modified-object-failed.csv",
            file_identifier,
            old_sha256,
            new_sha256,
        )


def handle_missing_object(
    github_url: str,
    sha256: str,
    metadata: Dict[str, Any],
    log_file: str = "handle-missing-object.csv",
) -> None:
    """Called when an object that is in Objaverse-XL is not found.

    Here, it is likely that the repository was deleted or renamed. If None, nothing
    will be done with the missing object.

    Args:
        github_url (str): GitHub URL of the 3D object.
        sha256 (str): SHA256 of the contents of the original 3D object.
        metadata (Dict[str, Any]): Metadata about the 3D object, including the GitHub
            organization and repo names.
        log_file (str): Name of the log file to save missing renders to.

    Returns:
        None
    """
    # log the missing object
    log_processed_object(log_file, github_url, sha256)


def get_example_objects() -> pd.DataFrame:
    """Returns a DataFrame of example objects to use for debugging."""
    return pd.read_json("example-objects.json", orient="records")


def render_github_objects(
    render_dir: str = "~/.objaverse",
    num_renders: int = 12,
    processes: Optional[int] = None,
    save_repo_format: Optional[Literal["zip", "tar", "tar.gz"]] = None,
    only_northern_hemisphere: bool = False,
    render_timeout: int = 300,
    gpu_devices: Optional[Union[int, List[int]]] = None,
) -> None:
    """Renders all GitHub objects in the Objaverse-XL dataset.

    Args:
        render_dir (str): Directory where the objects will be rendered.
        num_renders (int): Number of renders to save of the object.
        processes (Optional[int]): Number of processes to use for downloading the
            objects. If None, defaults to multiprocessing.cpu_count() * 3.
        save_repo_format (Optional[Literal["zip", "tar", "tar.gz"]]): If not None,
            the GitHub repo will be deleted after rendering each object from it.
        only_northern_hemisphere (bool): Only render the northern hemisphere of the
            object. Useful for rendering objects that are obtained from photogrammetry,
            since the southern hemisphere is often has holes.
        render_timeout (int): Number of seconds to wait for the rendering job to
            complete.
        gpu_devices (Optional[Union[int, List[int]]]): GPU device(s) to use for
            rendering. If an int, the GPU device will be randomly selected from 0 to
            gpu_devices - 1. If a list, the GPU device will be randomly selected from
            the list. If 0, the CPU will be used for rendering. If None, defaults to
            use all available GPUs.

    Returns:
        None
    """
    if platform.system() not in ["Linux", "Darwin"]:
        raise NotImplementedError(
            f"Platform {platform.system()} is not supported. Use Linux or MacOS."
        )

    # get the gpu devices to use
    parsed_gpu_devices: Union[int, List[int]] = 0
    if gpu_devices is None:
        parsed_gpu_devices = len(GPUtil.getGPUs())

    if processes is None:
        processes = multiprocessing.cpu_count() * 3

    objects = get_example_objects()
    download_github_objects(
        objects=objects,
        processes=processes,
        save_repo_format=save_repo_format,
        download_dir=render_dir,  # only used when save_repo_format is not None
        handle_found_object=partial(
            handle_found_object,
            render_dir=render_dir,
            num_renders=num_renders,
            only_northern_hemisphere=only_northern_hemisphere,
            gpu_devices=parsed_gpu_devices,
            render_timeout=render_timeout,
        ),
        handle_new_object=handle_new_object,
        handle_modified_object=partial(
            handle_modified_object,
            render_dir=render_dir,
            num_renders=num_renders,
            gpu_devices=parsed_gpu_devices,
            only_northern_hemisphere=only_northern_hemisphere,
        ),
        handle_missing_object=handle_missing_object,
    )


if __name__ == "__main__":
    fire.Fire(render_github_objects)