grounding_component.py

import argparse
from functools import partial
import cv2
import requests
import os
from io import BytesIO
from PIL import Image
import numpy as np
from pathlib import Path
import gradio as gr
import warnings
import torch
import cv2
import numpy as np

from groundingdino.models import build_model
from groundingdino.util.slconfig import SLConfig
from groundingdino.util.utils import clean_state_dict
from groundingdino.util.inference import annotate, load_image, predict
import groundingdino.datasets.transforms as T

from huggingface_hub import hf_hub_download

# Use this command for evaluate the GLIP-T model
config_file = "groundingdino/config/GroundingDINO_SwinT_OGC.py"
ckpt_repo_id = "ShilongLiu/GroundingDINO"
ckpt_filenmae = "groundingdino_swint_ogc.pth"

def load_model_hf(model_config_path, repo_id, filename, device='cpu'):
    args = SLConfig.fromfile(model_config_path)
    model = build_model(args)
    args.device = device

    cache_file = hf_hub_download(repo_id=repo_id, filename=filename)
    checkpoint = torch.load(cache_file, map_location='cpu')
    log = model.load_state_dict(clean_state_dict(checkpoint['model']), strict=False)
    print("Model loaded from {} \n => {}".format(cache_file, log))
    _ = model.eval()
    return model

def image_transform_grounding(init_image):
    transform = T.Compose([
        T.RandomResize([800], max_size=1333),
        T.ToTensor(),
        T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ])
    image, _ = transform(init_image, None)  # 3, h, w
    return init_image, image

def image_transform_grounding_for_vis(init_image):
    transform = T.Compose([
        T.RandomResize([800], max_size=1333),
    ])
    image, _ = transform(init_image, None)  # 3, h, w
    return image


model = load_model_hf(config_file, ckpt_repo_id, ckpt_filenmae)

def run_grounding(input_image, describe):
    pil_img = Image.fromarray(input_image)
    init_image = pil_img.convert("RGB")
    grounding_caption = describe
    box_threshold = 0.3
    text_threshold = 0.25

    _, image_tensor = image_transform_grounding(init_image)
    image_pil: Image = image_transform_grounding_for_vis(init_image)

    boxes, logits, phrases = predict(model, image_tensor, grounding_caption, box_threshold, text_threshold,
                                     device='cpu')
    annotated_frame = annotate(image_source=np.asarray(image_pil), boxes=boxes, logits=logits, phrases=phrases)
    image_with_box = Image.fromarray(cv2.cvtColor(annotated_frame, cv2.COLOR_BGR2RGB))
    return image_with_box