Classification and Segmentation Tests

Testing an image that shows multiple deer with various models.

Whole Deer

As far as I can tell the best result was achieved with Grounded-SAM.

Original Image

https://huggingface.co/spaces/IDEA-Research/Grounded-SAM (https://github.com/IDEA-Research/Grounded-Segment-Anything)

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

https://github.com/ultralytics/ultralytics (yolov8n.pt)

https://segment-anything.com/demo#

https://dinov2.metademolab.com/demos?category=segmentation

Deer Face Segmentation

https://github.com/IDEA-Research/Grounded-Segment-Anything

Running Grounded SAM locally in an attempt to focus on deer faces:

python grounded_sam_demo.py   \
    --config GroundingDINO/groundingdino/config/GroundingDINO_SwinT_OGC.py   \
    --grounded_checkpoint groundingdino_swint_ogc.pth   \
    --sam_checkpoint sam_vit_h_4b8939.pth   \
    --input_image ~/intro_thumbnail.jpg   \
    --output_dir "outputs"   \
    --box_threshold 0.3   \
    --text_threshold 0.25   \
    --text_prompt "deer face"   \
    --device "cuda"

Original image:

Grounded SAM output:

Mask: