dora/node-hub/dora-internvl/dora_internvl/main.py

"""TODO: Add docstring."""

import os

import numpy as np
import pyarrow as pa
import torch
import torchvision.transforms as T
from dora import Node
from PIL import Image
from torchvision.transforms.functional import InterpolationMode
from transformers import AutoModel, AutoTokenizer

IMAGENET_MEAN = (0.485, 0.456, 0.406)
IMAGENET_STD = (0.229, 0.224, 0.225)


def build_transform(input_size):
    """TODO: Add docstring."""
    MEAN, STD = IMAGENET_MEAN, IMAGENET_STD
    transform = T.Compose(
        [
            T.Lambda(lambda img: img.convert("RGB") if img.mode != "RGB" else img),
            T.Resize((input_size, input_size), interpolation=InterpolationMode.BICUBIC),
            T.ToTensor(),
            T.Normalize(mean=MEAN, std=STD),
        ],
    )
    return transform


def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size):
    """TODO: Add docstring."""
    best_ratio_diff = float("inf")
    best_ratio = (1, 1)
    area = width * height
    for ratio in target_ratios:
        target_aspect_ratio = ratio[0] / ratio[1]
        ratio_diff = abs(aspect_ratio - target_aspect_ratio)
        if ratio_diff < best_ratio_diff:
            best_ratio_diff = ratio_diff
            best_ratio = ratio
        elif ratio_diff == best_ratio_diff:
            if area > 0.5 * image_size * image_size * ratio[0] * ratio[1]:
                best_ratio = ratio
    return best_ratio


def dynamic_preprocess(
    image, min_num=1, max_num=12, image_size=448, use_thumbnail=False,
):
    """TODO: Add docstring."""
    orig_width, orig_height = image.size
    aspect_ratio = orig_width / orig_height

    # calculate the existing image aspect ratio
    target_ratios = set(
        (i, j)
        for n in range(min_num, max_num + 1)
        for i in range(1, n + 1)
        for j in range(1, n + 1)
        if i * j <= max_num and i * j >= min_num
    )
    target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])

    # find the closest aspect ratio to the target
    target_aspect_ratio = find_closest_aspect_ratio(
        aspect_ratio, target_ratios, orig_width, orig_height, image_size,
    )

    # calculate the target width and height
    target_width = image_size * target_aspect_ratio[0]
    target_height = image_size * target_aspect_ratio[1]
    blocks = target_aspect_ratio[0] * target_aspect_ratio[1]

    # resize the image
    resized_img = image.resize((target_width, target_height))
    processed_images = []
    for i in range(blocks):
        box = (
            (i % (target_width // image_size)) * image_size,
            (i // (target_width // image_size)) * image_size,
            ((i % (target_width // image_size)) + 1) * image_size,
            ((i // (target_width // image_size)) + 1) * image_size,
        )
        # split the image
        split_img = resized_img.crop(box)
        processed_images.append(split_img)
    assert len(processed_images) == blocks
    if use_thumbnail and len(processed_images) != 1:
        thumbnail_img = image.resize((image_size, image_size))
        processed_images.append(thumbnail_img)
    return processed_images


def load_image(image_array: np.array, input_size=448, max_num=12):
    """TODO: Add docstring."""
    image = Image.fromarray(image_array).convert("RGB")
    transform = build_transform(input_size=input_size)
    images = dynamic_preprocess(
        image, image_size=input_size, use_thumbnail=True, max_num=max_num,
    )
    pixel_values = [transform(image) for image in images]
    pixel_values = torch.stack(pixel_values)
    return pixel_values


def main():
    # Handle dynamic nodes, ask for the name of the node in the dataflow, and the same values as the ENV variables.
    """TODO: Add docstring."""
    model_path = os.getenv("MODEL", "OpenGVLab/InternVL2-1B")
    device = "cuda:0" if torch.cuda.is_available() else "cpu"

    # If you want to load a model using multiple GPUs, please refer to the `Multiple GPUs` section.
    model = (
        AutoModel.from_pretrained(
            model_path,
            torch_dtype=torch.bfloat16,
            low_cpu_mem_usage=True,
            use_flash_attn=True,
            trust_remote_code=True,
        )
        .eval()
        .to(device)
    )
    tokenizer = AutoTokenizer.from_pretrained(
        model_path, trust_remote_code=True, use_fast=False,
    )

    node = Node()

    question = "<image>\nPlease describe the image shortly."
    frame = None
    pa.array([])  # initialize pyarrow array

    for event in node:
        event_type = event["type"]

        if event_type == "INPUT":
            event_id = event["id"]

            if event_id == "image":
                storage = event["value"]
                metadata = event["metadata"]
                encoding = metadata["encoding"]
                width = metadata["width"]
                height = metadata["height"]

                if encoding == "bgr8" or encoding == "rgb8":
                    channels = 3
                    storage_type = np.uint8
                else:
                    raise RuntimeError(f"Unsupported image encoding: {encoding}")

                frame = (
                    storage.to_numpy()
                    .astype(storage_type)
                    .reshape((height, width, channels))
                )
                if encoding == "bgr8":
                    frame = frame[:, :, ::-1]  # OpenCV image (BGR to RGB)
                elif encoding == "rgb8":
                    pass
                else:
                    raise RuntimeError(f"Unsupported image encoding: {encoding}")

            elif event_id == "text":
                question = "<image>\n" + event["value"][0].as_py()
                if frame is not None:
                    # set the max number of tiles in `max_num`
                    pixel_values = (
                        load_image(frame, max_num=12).to(torch.bfloat16).cuda()
                    )
                    generation_config = dict(max_new_tokens=1024, do_sample=True)
                    response = model.chat(
                        tokenizer, pixel_values, question, generation_config,
                    )
                    node.send_output(
                        "text",
                        pa.array([response]),
                        metadata,
                    )

        elif event_type == "ERROR":
            raise RuntimeError(event["error"])


if __name__ == "__main__":
    main()