lc_GroundingDINO

This package exposes GroundingDINO with local_comm. This allows GroundingDINO to be isolated into a conda environment and called from another with minimal dependancies.

Install the server

1. Clone repository (make sure to use the --recursive flag)
2. Change directory: cd lc_GroundingDINO
3. Create conda environment: conda create -n lc_groundingdino python=3.10 (only tested with Python 3.10)
4. Activate conda environment: conda activate lc_groundingdino
5. Install dependencies

• GroundingDINO: cd extern/GroundingDINO and follow instructions, here (note, this may be tricky).
• Change directory: cd /path/to/lc_GroundingDINO
• local_comm: pip install extern/local_comm

6. Install lc_groundingdino: pip install .

You can start the server using

conda activate lc_groundingdino
python -m lc_groundingdino.image_inference_server

Install the client

Note, the client will require lc_groundingdino to be installed but this does not require GroundingDINO to be installed. Check pyproject.toml for the minimal list of dependancies (note, local_comm is dependency-free).

1. Change directory: cd /path/to/lc_GroundingDINO
2. Create and activate your conda environment, for example called lc_groundingdino_client.
3. Install local_comm: pip install extern/local_comm
4. Install lc_groundingdino: pip install .

Examples

Basic example

Run the following basic example to check that GroundingDINO is working.

conda activate lc_groundingdino
python scripts/basic_example.py /path/to/image_filename PROMPT1 PROMPT2

Server/client example

Open a terminal and start the server.

conda activate lc_groundingdino
python -m lc_groundingdino.image_inference_server

Open another terminal and test the client.

conda activate lc_groundingdino_client
python scripts/test_client.py scripts/image.jpg person containers bowl stove

You should expect to see the following output.

[info] Sending request on channel 'groundingdino'
[info] Call took 0.192s (5.20 Hz)
[info] Received 15 detections
  00: person        logit=0.912  box=[0.17006513476371765, 0.5478484034538269, 0.33907875418663025, 0.9008976221084595]
  01: bowl          logit=0.673  box=[0.43435776233673096, 0.7306913137435913, 0.13680307567119598, 0.1269308179616928]
  02: containers    logit=0.409  box=[0.37393122911453247, 0.08947072178125381, 0.03197794407606125, 0.09651652723550797]
  03: containers    logit=0.410  box=[0.3437686860561371, 0.08994346857070923, 0.03408919647336006, 0.0959039255976677]
  04: containers    logit=0.418  box=[0.2570761442184448, 0.09374231100082397, 0.029447609558701515, 0.08323606848716736]
  05: stove         logit=0.404  box=[0.8802738785743713, 0.8256462216377258, 0.23787285387516022, 0.34573885798454285]
  06: containers    logit=0.368  box=[0.7540533542633057, 0.09397221356630325, 0.05794242024421692, 0.18710771203041077]
  07: containers    logit=0.358  box=[0.5850163102149963, 0.09245003759860992, 0.05671674385666847, 0.18306756019592285]
  08: containers    logit=0.351  box=[0.7260599732398987, 0.28970515727996826, 0.03193245455622673, 0.11721739917993546]
  09: containers    logit=0.367  box=[0.7578036189079285, 0.3053896427154541, 0.03669160604476929, 0.08268167823553085]
  10: containers    logit=0.354  box=[0.47004619240760803, 0.09349002689123154, 0.05757146328687668, 0.18494975566864014]
  11: containers    logit=0.398  box=[0.2103295773267746, 0.09442437440156937, 0.06749828904867172, 0.08302118629217148]
  12: containers    logit=0.365  box=[0.6964897513389587, 0.0910215899348259, 0.054258622229099274, 0.18105098605155945]
  13: containers    logit=0.364  box=[0.5256297588348389, 0.0929589495062828, 0.05548851564526558, 0.1840696483850479]
  14: containers    logit=0.364  box=[0.6415461897850037, 0.09309326112270355, 0.0557854026556015, 0.1850593537092209]
[info] Wrote annotated image to test_output.png

The test_output.png should be the same as the figure above.