third_parts/recognize_anything/build_ram_plus.py

import argparse
import numpy as np
import random
import os

import torch

from PIL import Image
from huggingface_hub import hf_hub_download

from .ram.models import ram_plus
from .ram import get_transform


def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size):
    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=6, image_size=448, use_thumbnail=False):
    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_file, input_size=384, max_num=12, upscale=False):
    image = Image.open(image_file)
    if upscale:
        image = image.resize((image.width * 2, image.height * 2), Image.BILINEAR)
    images = dynamic_preprocess(image, image_size=input_size, use_thumbnail=True, max_num=max_num)
    return images


class RAMPredictor(object):
    def __init__(
            self,
            repo_id="xinyu1205/recognize-anything-plus-model",
            checkpoint_file="ram_plus_swin_large_14m.pth",
            image_size=384,
            device='cuda',
        ):
        super().__init__()

        self.image_size = image_size
        self.device = device
        self.transform = get_transform(image_size=image_size)

        if os.path.exists(checkpoint_file):
            init_checkpoint = checkpoint_file
        else:
            init_checkpoint = hf_hub_download(repo_id=repo_id, filename=checkpoint_file)
        
        self.model = ram_plus(pretrained=init_checkpoint, image_size=image_size, vit='swin_l', text_encoder_type="third_parts/recognize_anything/google-bert/bert-base-uncased")
        self.model.eval()
        self.model.to(device)

    def run_on_image(
            self,
            image_file_path,
            dynamic_resolution=False,
        ):
        if dynamic_resolution:
            images = load_image(image_file_path, input_size=self.image_size)
            images = [self.transform(image) for image in images]
            images = torch.stack(images).to(self.device)
        else:
            if not isinstance(image_file_path, str):
                images = self.transform(image_file_path).unsqueeze(0).to(self.device)
            else:
                images = self.transform(Image.open(image_file_path)).unsqueeze(0).to(self.device)

        res = self.model.generate_tag(images)

        return res
    

def build_ram_predictor(override_ckpt_file="", device="cuda"):
    repo_id="xinyu1205/recognize-anything-plus-model"
    checkpoint_file="ram_plus_swin_large_14m.pth"

    if os.path.exists(override_ckpt_file):
        checkpoint_file = override_ckpt_file
    
    ram_predictor = RAMPredictor(repo_id, checkpoint_file, image_size=384, device=device)

    return ram_predictor

if __name__ == "__main__":
    ram_predictor = build_ram_predictor(override_ckpt_file="xinyu1205/recognize-anything-plus-model/ram_plus_swin_large_14m.pth")

    res = ram_predictor.run_on_image(image_file_path="sa_7963505.jpg", dynamic_resolution=True)

    tag_list = []
    for tag_string in res[0]:
        tags = tag_string.split(' | ')
        tag_list += tags
    tags = list(set(tag_list))

    print("Image Tags: ", tags)