GroundingDINO-Inference

This project represents a GroundingDINO Inference (zero-shot object detection) procedure with both methods (CLI and Script). This implementation will help the reader to know the sequence of commands and exemplifying commands for running a quick zero-shot object detection. Additionally, the reader may get insight into code (script) execution. This is the Google Colab implementation and has some prerequisites for the complete execution of the script.

1. Colab Notebook for inference on CPU

2. Colab Notebook for inference on GPU

Prerequisite for the following Colab notebooks: The user would need a folder that should be uploaded on the drive. The link to the folder is https://drive.google.com/drive/folders/1ctzsamgcgQ9OpxclnNlj7iLYsbbbh2R3?usp=share_link. The user needs to download that folder and upload it to the drive.

3. Colab Notebook for inference using Drive folder

4. Colab Notebook for inference using Drive folder

Table Of Contents

1. Demo Results
2. Inference Steps (CLI):
3. Inference (Code):

Demo Results

Demo1

Demo2

Demo3

Demo4

Demo5

Inference Steps (CLI): [GroundingDINO_Trial1.ipynb]

Step1

from google.colab import drive
drive.mount('/content/drive')

Step2

cd drive/MyDrive/GroundingDINO/

Step3

pip install -q -e .

Step4

python demo/inference_on_a_image.py -c groundingdino/config/GroundingDINO_SwinT_OGC.py -p weights/groundingdino_swint_ogc.pth -i frame1.jpg -o "output" -t "bike . person . car ." 

Results Visualization (CLI)

Go to the output folder that would be created automatically.

Inference (Code): [GroundingDINO_Trial2.ipynb]

from groundingdino.util.inference import load_model, load_image, predict, annotate
import cv2

model = load_model("groundingdino/config/GroundingDINO_SwinT_OGC.py", "weights/groundingdino_swint_ogc.pth")
IMAGE_PATH = "im1.jpg"
TEXT_PROMPT = "persons . sofas . fans"
BOX_TRESHOLD = 0.35
TEXT_TRESHOLD = 0.25

image_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("annotated_image.jpg", annotated_frame)

Results Visualization (Code)

An annotated image will be created in the code directory.

References

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

Important Announcement

This is just the beginning of the project. More options and features may be provided in the future.