OV_GroundedSAM

Run Grounded SAM with OpenVINO on Intel dGPU.

To convert models

GroundingDINO

Check https://github.com/wenyi5608/GroundingDINO for more details.

SAM

pip install -e git+https://github.com/facebookresearch/segment-anything.git#egg=segment_anything
python convert_sam.py vit_l models/sam

ViT

pip install -e git+https://github.com/lukemelas/PyTorch-Pretrained-ViT.git#egg=pytorch_pretrained_vit
python convert_vit.py L_32 -o models/vit

To benchmark models

GroundingDINO

benchmark_app -m models/dino/groundingdino_512.xml -d GPU -data_shape img[1,3,512,512],input_ids[1,6],attention_mask[1,6],position_ids[1,6],token_type_ids[1,6],text_token_mask[1,6,6] -hint latency

SAM

benchmark_app -m models/sam/sam_image_encoder_vit_l.xml -hint latency -d GPU
benchmark_app -m models/sam/sam_mask_predictor_vit_b.xml -d GPU -data_shape "[1,256,64,64]" -hint latency

ViT

benchmark_app vit/vit_b32_fp16.xml -hint latency -d GPU