GroundingDINO/demo/inference_on_a_image.py

import argparse
import os
import sys

import numpy as np
import torch
from PIL import Image, ImageDraw, ImageFont

import groundingdino.datasets.transforms as T
from groundingdino.models import build_model
from groundingdino.util import box_ops
from groundingdino.util.slconfig import SLConfig
from groundingdino.util.utils import clean_state_dict, get_phrases_from_posmap


def plot_boxes_to_image(image_pil, tgt):
    H, W = tgt["size"]
    boxes = tgt["boxes"]
    labels = tgt["labels"]
    assert len(boxes) == len(labels), "boxes and labels must have same length"

    draw = ImageDraw.Draw(image_pil)
    mask = Image.new("L", image_pil.size, 0)
    mask_draw = ImageDraw.Draw(mask)

    # draw boxes and masks
    for box, label in zip(boxes, labels):
        # from 0..1 to 0..W, 0..H
        box = box * torch.Tensor([W, H, W, H])
        # from xywh to xyxy
        box[:2] -= box[2:] / 2
        box[2:] += box[:2]
        # random color
        color = tuple(np.random.randint(0, 255, size=3).tolist())
        # draw
        x0, y0, x1, y1 = box
        x0, y0, x1, y1 = int(x0), int(y0), int(x1), int(y1)

        draw.rectangle([x0, y0, x1, y1], outline=color, width=6)
        # draw.text((x0, y0), str(label), fill=color)

        font = ImageFont.load_default()
        if hasattr(font, "getbbox"):
            bbox = draw.textbbox((x0, y0), str(label), font)
        else:
            w, h = draw.textsize(str(label), font)
            bbox = (x0, y0, w + x0, y0 + h)
        # bbox = draw.textbbox((x0, y0), str(label))
        draw.rectangle(bbox, fill=color)
        draw.text((x0, y0), str(label), fill="white")

        mask_draw.rectangle([x0, y0, x1, y1], fill=255, width=6)

    return image_pil, mask


def load_image(image_path):
    # load image
    image_pil = Image.open(image_path).convert("RGB")  # load 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(image_pil, None)  # 3, h, w
    return image_pil, image


def load_model(model_config_path, model_checkpoint_path, cpu_only=False):
    args = SLConfig.fromfile(model_config_path)
    args.device = "cuda" if not cpu_only else "cpu"
    model = build_model(args)
    checkpoint = torch.load(model_checkpoint_path, map_location="cpu")
    load_res = model.load_state_dict(clean_state_dict(checkpoint["model"]), strict=False)
    print(load_res)
    _ = model.eval()
    return model


def get_grounding_output(model, image, caption, box_threshold, text_threshold, with_logits=True, cpu_only=False):
    caption = caption.lower()
    caption = caption.strip()
    if not caption.endswith("."):
        caption = caption + "."
    device = "cuda" if not cpu_only else "cpu"
    model = model.to(device)
    image = image.to(device)
    with torch.no_grad():
        outputs = model(image[None], captions=[caption])
    logits = outputs["pred_logits"].cpu().sigmoid()[0]  # (nq, 256)
    boxes = outputs["pred_boxes"].cpu()[0]  # (nq, 4)
    logits.shape[0]

    # filter output
    logits_filt = logits.clone()
    boxes_filt = boxes.clone()
    filt_mask = logits_filt.max(dim=1)[0] > box_threshold
    logits_filt = logits_filt[filt_mask]  # num_filt, 256
    boxes_filt = boxes_filt[filt_mask]  # num_filt, 4
    logits_filt.shape[0]

    # get phrase
    tokenlizer = model.tokenizer
    tokenized = tokenlizer(caption)
    # build pred
    pred_phrases = []
    for logit, box in zip(logits_filt, boxes_filt):
        pred_phrase = get_phrases_from_posmap(logit > text_threshold, tokenized, tokenlizer)
        if with_logits:
            pred_phrases.append(pred_phrase + f"({str(logit.max().item())[:4]})")
        else:
            pred_phrases.append(pred_phrase)

    return boxes_filt, pred_phrases


if __name__ == "__main__":

    parser = argparse.ArgumentParser("Grounding DINO example", add_help=True)
    parser.add_argument("--config_file", "-c", type=str, required=True, help="path to config file")
    parser.add_argument(
        "--checkpoint_path", "-p", type=str, required=True, help="path to checkpoint file"
    )
    parser.add_argument("--image_path", "-i", type=str, required=True, help="path to image file")
    parser.add_argument("--text_prompt", "-t", type=str, required=True, help="text prompt")
    parser.add_argument(
        "--output_dir", "-o", type=str, default="outputs", required=True, help="output directory"
    )

    parser.add_argument("--box_threshold", type=float, default=0.3, help="box threshold")
    parser.add_argument("--text_threshold", type=float, default=0.25, help="text threshold")

    parser.add_argument("--cpu-only", action="store_true", help="running on cpu only!, default=False")
    args = parser.parse_args()

    # cfg
    config_file = args.config_file  # change the path of the model config file
    checkpoint_path = args.checkpoint_path  # change the path of the model
    image_path = args.image_path
    text_prompt = args.text_prompt
    output_dir = args.output_dir
    box_threshold = args.box_threshold
    text_threshold = args.box_threshold

    # make dir
    os.makedirs(output_dir, exist_ok=True)
    # load image
    image_pil, image = load_image(image_path)
    # load model
    model = load_model(config_file, checkpoint_path, cpu_only=args.cpu_only)

    # visualize raw image
    image_pil.save(os.path.join(output_dir, "raw_image.jpg"))

    # run model
    boxes_filt, pred_phrases = get_grounding_output(
        model, image, text_prompt, box_threshold, text_threshold, cpu_only=args.cpu_only
    )

    # visualize pred
    size = image_pil.size
    pred_dict = {
        "boxes": boxes_filt,
        "size": [size[1], size[0]],  # H,W
        "labels": pred_phrases,
    }
    # import ipdb; ipdb.set_trace()
    image_with_box = plot_boxes_to_image(image_pil, pred_dict)[0]
    image_with_box.save(os.path.join(output_dir, "pred.jpg"))