2Phuong5Nam4/SAM3_carton_detection

SAM3 LoRA for Carton Detection

This is a LoRA adapter for SAM3 fine-tuned for carton detection.

Usage

from sam3.model_builder import build_sam3_image_model
from lora_layers import LoRAConfig, apply_lora_to_model, load_lora_weights
from huggingface_hub import hf_hub_download

# Build base model
model = build_sam3_image_model(device='cuda', load_from_HF=True)

# Apply LoRA structure  
lora_config = LoRAConfig(rank=16, alpha=32, dropout=0.0)
model = apply_lora_to_model(model, lora_config)

# Download and load weights
weights_path = hf_hub_download(repo_id="YOUR_USERNAME/SAM3_carton_detection", filename="lora_weights.pt")
load_lora_weights(model, weights_path)

model.eval()
## Training Details
Rank: 16
Alpha: 32
Task: Instance segmentation of cartons