### Run Quickstart Script

Source: https://github.com/joey-s-liu/medsam3/blob/main/PROJECT_STRUCTURE.txt

Execute the main quickstart script to set up the project environment.

```bash
$ ./quickstart.sh
```

--------------------------------

### Install MedSAM3 Dependencies

Source: https://github.com/joey-s-liu/medsam3/blob/main/README.md

Clone the MedSAM3 repository and install its dependencies. Ensure you log in to Hugging Face with your token.

```bash
git clone https://github.com/Joey-S-Liu/MedSAM3.git
cd MedSAM3

pip install -e .

hf auth login
# Paste your token when prompted
```

--------------------------------

### Create Output Directory for Data Preparation

Source: https://github.com/joey-s-liu/medsam3/blob/main/PROJECT_STRUCTURE.txt

Use the prepare_data.py script to create the output directory for processed data.

```python
python prepare_data.py create --output_dir data
```

--------------------------------

### Train MedSAM3 Model

Source: https://github.com/joey-s-liu/medsam3/blob/main/README.md

Initiate the training process for the MedSAM3 model using a specified configuration file.

```bash
python3 train_sam3_lora_native.py --config configs/full_lora_config.yaml
```

--------------------------------

### Run Inference with LoRA Weights and Text Prompt

Source: https://github.com/joey-s-liu/medsam3/blob/main/PROJECT_STRUCTURE.txt

Perform inference using trained LoRA weights. Provide the path to the LoRA weights, the image to process, and a text prompt.

```python
python inference.py \
    --lora_weights outputs/sam3_lora/best_model/lora_weights.pt \
    --image test.jpg \
    --text_prompt "yellow school bus"
```

--------------------------------

### Train MedSAM3 Model with LoRA

Source: https://github.com/joey-s-liu/medsam3/blob/main/PROJECT_STRUCTURE.txt

Initiate the training process for the MedSAM3 model using LoRA configuration. Specify the configuration file path.

```python
python train_sam3_lora.py --config configs/base_config.yaml
```

--------------------------------

### Run Inference with MedSAM3

Source: https://github.com/joey-s-liu/medsam3/blob/main/README.md

Execute the inference script with specified configuration, image, prompt, and output settings. Adjust threshold and nms-iou for optimal results based on modality and target.

```bash
python3 infer_sam.py \
  --config configs/full_lora_config.yaml \
  --image path/to/image.jpg \
  --prompt "skin lesion" \
  --threshold 0.5 \
  --nms-iou 0.5 \
  --output skin_lesion.png
```

--------------------------------

### Prepare COCO JSON Data

Source: https://github.com/joey-s-liu/medsam3/blob/main/PROJECT_STRUCTURE.txt

Process COCO JSON annotations and images to prepare data for training. Specify the COCO JSON file, image directory, and output directory.

```python
python prepare_data.py coco --coco_json annotations.json \
                                  --images_dir images/ \
                                  --output_dir data
```

--------------------------------

### MedSAM3 Project Citation

Source: https://github.com/joey-s-liu/medsam3/blob/main/README.md

BibTeX entry for citing the MedSAM3 project. Include this in your research papers if the project is utilized.

```bibtex
@misc{liu2025medsam3delvingsegmentmedical,
      title={MedSAM3: Delving into Segment Anything with Medical Concepts},
      author={Anglin Liu and Rundong Xue and Xu R. Cao and Yifan Shen and Yi Lu and Xiang Li and Qianqian Chen and Jintai Chen},
      year={2025},
      eprint={2511.19046},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2511.19046},
}
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.