grounding-DINO 실습 with python!

🦖 (English) Grounding DINO Practice! Running the Evolved Model of DINO

This post is a hands-on guide to the Grounding DINO, the evolved version of DINO!
Just like DINO, we clone the model from GitHub and run it — and surprisingly, it’s even simpler 😊
So this time, we’ll skip the theory for now and jump straight into running the code!!

---

🧱 1. Clone the GitHub Repository

\($ git clone https://github.com/IDEA-Research/GroundingDINO.git cd GroundingDINO/\)$

---

📦 2. Install the Model

\($ pip install -e .\)$

---

🧊 3. Download the Pretrained Weights

\($ mkdir weights cd weights/ wget -q https://github.com/IDEA-Research/GroundingDINO/releases/download/v0.1.0-alpha/groundingdino_swint_ogc.pth cd ..\)$

---

🚀 4. Run the Inference (Command Template)

The command structure for running Grounding DINO is shown below.
Each part marked with {} can be replaced with the appropriate value for your environment:

\($ CUDA_VISIBLE_DEVICES={GPU_ID} python demo/inference_on_a_image.py \ -c {CONFIG_PATH} \ -p {CHECKPOINT_PATH} \ -i {INPUT_IMAGE_PATH} \ -o {OUTPUT_DIR} \ -t "{TEXT_PROMPT}"\)$

Example values:

Variable	Description
{GPU_ID}	GPU ID to use (e.g., 0) — you can check it using nvidia-smi
{CONFIG_PATH}	Config file path (e.g., groundingdino/config/GroundingDINO_SwinT_OGC.py)
{CHECKPOINT_PATH}	Path to the pretrained weights (e.g., weights/groundingdino_swint_ogc.pth)
{INPUT_IMAGE_PATH}	Input image path (e.g., /home/user/images/catch_rugby_ball_001480.jpg)
{OUTPUT_DIR}	Directory to save the result (e.g., /home/user/images/GroundingDINO/results)
{TEXT_PROMPT}	Text prompt to detect (e.g., "chair")

• You can change the TEXT_PROMPT to try different detection phrases!!

---

✅ Real-World Prompt Tests!

Let’s now try running the inference while changing only the TEXT_PROMPT and see the results!

---

person — A common COCO category.

Of course it works great!

---

cat — Will it produce any false positives even when the object isn’t there?

Nothing detected! Well done 😎

---

rugby — Likely missing from most test sets. Will it still work?

Oh nice! It makes sense!!

---

jump — Now let’s try a verb!

Whoa~ It works for actions too!?

---

player is jumping — What about a full sentence?

Hmm… Seems like it breaks it apart instead of treating it as one phrase.

---

🎉 Conclusion

Grounding DINO was very easy to install, and the inference workflow is intuitive!
Especially useful if you want to experiment with diverse text prompts.
If it could even understand full sentences, and support segmentation too —
that would be amazing, right!? 😄

Well, maybe there’s a model out there that already does that?
Let’s keep exploring together!

---

🦖(한국어) Grounding DINO 실습! DINO의 진화 모델을 직접 실행해보자!

이번 포스팅은 DINO의 후속 모델인 Grounding DINO 실습입니다!
DINO와 마찬가지로 GitHub repo에서 코드를 내려받아 실행하지만, 오히려 더 간단하게 구성되어 있더라구요 😊
그래서 이번엔 이론은 잠시 뒤로 미루고, 바로 코드부터 실행해봅니다!!

---

🧱 1. GitHub 저장소 클론

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

---

📦 2. 모델 설치

pip install -e .

---

🧊 3. Pretrained Weight 다운로드

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

---

🚀 4. 객체 탐지 실험 실행 (명령어 템플릿)

다음 명령어는 Grounding DINO를 실행하는 전체 구조입니다.
필요한 값들은 {} 변수로 구성되어 있어 상황에 맞게 대체하면 됩니다:

CUDA_VISIBLE_DEVICES={GPU_ID} python demo/inference_on_a_image.py \
  -c {CONFIG_PATH} \
  -p {CHECKPOINT_PATH} \
  -i {INPUT_IMAGE_PATH} \
  -o {OUTPUT_DIR} \
  -t "{TEXT_PROMPT}"

예시 값 대입:

변수명	설명
{GPU_ID}	사용할 GPU ID (예: 0) - nvidia-smi 하면 확인할수 있습니다!!
{CONFIG_PATH}	설정 파일 경로 (예: groundingdino/config/GroundingDINO_SwinT_OGC.py)
{CHECKPOINT_PATH}	모델 가중치 경로 (예: weights/groundingdino_swint_ogc.pth)
{INPUT_IMAGE_PATH}	감지할 이미지 경로 (예: /home/user/images/catch_rugby_ball_001480.jpg)
{OUTPUT_DIR}	출력 저장 디렉터리 (예: /home/user/images/GroundingDINO/results)
{TEXT_PROMPT}	탐지할 대상 텍스트 (예: "chair")

• 여기서 TEXT_PROMPT 에 다양한 값을 넣으며 테스트해볼 수 있습니다!!

✅ 실전 프롬포트별 테스트!!

이제 TEXT_PROMPT 를 바꿔가며 결과를 보겠습니다!!

person. 가장 간단하며 기존 coco dataset에 있는 person!!

역시 잘하는구만~~

cat. 없는것을 오탐지하지는 않을까요!?

아무것도 안잡네? 잘했어!!

테스트 셋에도 없었을 rugby. 럭비라는 단어로 작동을 할까요!?

럭비~~ 그럴듯해 좋아!?!

jump 이번엔 동사으로!!

오~ 동사 잘하는걸!?!

player is jumping 이번엔 문장!!

아쉽게도 문장을 하나로 인식하는게 아니라 끊어서 보는듯하네요!!

---

🎉 마무리

Grounding DINO는 설치도 간단하고, inference도 직관적으로 되어 있어서 바로 실험해보기 좋았습니다!
특히, 다양한 텍스트 프롬프트를 실험해보고 싶을 때 매우 유용할 것 같아요.
문장까지 잘 인지하고, Segment 까지할수 있다면!!
얼~~마나 좋을까요!!
그런데! 그런 연구도 있지 않을까요!?
함께 공부해봅시다!^^