Models
LLaVA vs. DETIC

LLaVA vs. DETIC

Both LLaVA-1.5 and DETIC are commonly used in computer vision projects. Below, we compare and contrast LLaVA-1.5 and DETIC.

Models

icon-model

LLaVA-1.5

LLaVA is an open source multimodal language model that you can use for visual question answering and has limited support for object detection.
icon-model

DETIC

Detic is an open source segmentation model developed by Meta Research and released in 2022.
Model Type
Object Detection
--
Instance Segmentation
--
Model Features
Item 1 Info
Item 2 Info
Architecture
--
--
Frameworks
--
PyTorch
--
Annotation Format
Instance Segmentation
Instance Segmentation
GitHub Stars
16,000
--
1.8k+
--
License
Apache-2.0
--
Apache 2.0
--
Training Notebook

Compare LLaVA-1.5 and DETIC with Autodistill

Using Autodistill, you can compare LLaVA and DETIC on your own images in a few lines of code.

Here is an example comparison:

To start a comparison, first install the required dependencies:


pip install autodistill autodistill-llava autodistill-detic

Next, create a new Python file and add the following code:


from autodistill_llava import LLaVA
from autodistill_detic import DETIC

from autodistill.detection import CaptionOntology
from autodistill.utils import compare

ontology = CaptionOntology(
    {
        "solar panel": "solar panel",
    }
)

models = [
    LLaVA(ontology=ontology),
    DETIC(ontology=ontology)
]

images = [
    "/home/user/autodistill/solarpanel1.jpg",
    "/home/user/autodistill/solarpanel2.jpg"
]

compare(
    models=models,
    images=images
)

Above, replace the images in the `images` directory with the images you want to use.

The images must be absolute paths.

Then, run the script.

You should see a model comparison like this:

When you have chosen a model that works best for your use case, you can auto label a folder of images using the following code:


base_model.label(
  input_folder="./images",
  output_folder="./dataset",
  extension=".jpg"
)

Compare LLaVA vs. DETIC

Provide your own image below to test YOLOv8 and YOLOv9 model checkpoints trained on the Microsoft COCO dataset.

COCO can detect 80 common objects, including cats, cell phones, and cars.