“Grounding DINO：Marrying DINO with Grounded Pre-Training for Open-Set Object Detection”的官方 PyTorch 实现：SoTA 开放集对象检测器。

一、Helpful Tutorial

论文地址：

https://arxiv.org/abs/2303.05499

在 YouTube 上观看介绍视频：

https://www.youtube.com/watch?v=wxWDt5UiwY8&feature=youtu.be

Try the Colab Demo：

https://colab.research.google.com/github/roboflow-ai/notebooks/blob/main/notebooks/zero-shot-object-detection-with-grounding-dino.ipynb

Try Official Huggingface Demo：

https://huggingface.co/spaces/ShilongLiu/Grounding_DINO_demo

二、相关的论文工作

2.1 相关的论文整理

• Grounded-SAM: Marrying Grounding DINO with Segment Anything
• Grounding DINO with Stable Diffusion
• Grounding DINO with GLIGEN for Controllable Image Editing
• OpenSeeD: A Simple and Strong Openset Segmentation Model
• SEEM: Segment Everything Everywhere All at Once
• X-GPT: Conversational Visual Agent supported by X-Decoder
• GLIGEN: Open-Set Grounded Text-to-Image Generation
• LLaVA: Large Language and Vision Assistant

2.2 论文的亮点

本工作的亮点：

1. Open-Set Detection. Detect everything with language!
2. High Performancce. COCO zero-shot 52.5 AP (training without COCO data!). COCO fine-tune 63.0 AP.
3. Flexible. Collaboration with Stable Diffusion for Image Editting.

2.3 论文介绍

2.4 Marrying Grounding DINO and GLIGEN

2.5 输入和输出的说明 / 提示

• Grounding DINO accepts an (image, text) pair as inputs.
• It outputs 900 (by default) object boxes. Each box has similarity scores across all input words. (as shown in Figures below.)
• We defaultly choose the boxes whose highest similarities are higher than a box_threshold.
• We extract the words whose similarities are higher than the text_threshold as predicted labels.
• If you want to obtain objects of specific phrases, like the dogs in the sentence two dogs with a stick., you can select the boxes with highest text similarities with dogs as final outputs.
• Note that each word can be split to more than one tokens with differetn tokenlizers. The number of words in a sentence may not equal to the number of text tokens.
• We suggest separating different category names with . for Grounding DINO.
  
  

三、环境配置过程

3.1 我的环境

系统：最新的ubuntu系统

显卡：3090

CUDA：11.3

如果您有 CUDA 环境，请确保设置了环境变量 CUDA_HOME。 如果没有可用的 CUDA，它将在 CPU-only 模式下编译。

3.2 配置过程

3.2.1 Clone the GroundingDINO repository from GitHub

git clone https://github.com/IDEA-Research/GroundingDINO.git

下载后即可找到对应的文件夹：

3.2.2 Change the current directory to the GroundingDINO folder

cd GroundingDINO/

3.2.3 Install the required dependencies in the current directory

pip3 install -q -e .

不知道为什么，我这个下载一直报错！换一个新的下载方式：

python setup.py install

但是也会飘红！

这个时候不要害怕，遇到错误的包，直接使用 pip 下载即可，耐得住性子，最后再运行上面的安装命令，即可顺利成功！

3.2.4 Create a new directory called “weights” to store the model weights

mkdir weights

Change the current directory to the “weights” folder：

cd weights

Download the model weights file：

wget -q https://github.com/IDEA-Research/GroundingDINO/releases/download/v0.1.0-alpha/groundingdino_swint_ogc.pth

四、测试

Check your GPU ID (only if you’re using a GPU)：

nvidia-smi

Replace {GPU ID}, image_you_want_to_detect.jpg, and “dir you want to save the output” with appropriate values in the following command：

CUDA_VISIBLE_DEVICES={GPU ID} python demo/inference_on_a_image.py \
-c /GroundingDINO/groundingdino/config/GroundingDINO_SwinT_OGC.py \
-p /GroundingDINO/weights/groundingdino_swint_ogc.pth \
-i image_you_want_to_detect.jpg \
-o "dir you want to save the output" \
-t "chair"[--cpu-only] # open it for cpu mode

当然了，我们也可以使用 Python 进行测试：

from groundingdino.util.inference import load_model, load_image, predict, annotate
import cv2model = load_model("./GroundingDINO/groundingdino/config/GroundingDINO_SwinT_OGC.py", "./GroundingDINO/weights/groundingdino_swint_ogc.pth")
IMAGE_PATH = "./GroundingDINO/weights/1.png"
TEXT_PROMPT = "person . bike . bottle ."
BOX_TRESHOLD = 0.35
TEXT_TRESHOLD = 0.25image_source, image = load_image(IMAGE_PATH)boxes, logits, phrases = predict(model=model,image=image,caption=TEXT_PROMPT,box_threshold=BOX_TRESHOLD,text_threshold=TEXT_TRESHOLD
)annotated_frame = annotate(image_source=image_source, boxes=boxes, logits=logits, phrases=phrases)
cv2.imwrite("./GroundingDINO/weights/annotated_image.jpg", annotated_frame)

我们的测试原图片为：

测试后的图片为：