DINO Pipeline

This repository contains a pipeline that utilizes DINO (Vision Transformer) for processing and inference tasks. It includes preprocessing and inference components, with Docker support for streamlined execution in isolated environments.

Features

• Preprocessing: Prepares the data for inference.
• Inference: Utilizes DINO for visual inference and attention inspection.
• Docker Support: Run the pipeline in an isolated Docker environment for consistency and ease of use.

Repository Structure

.
├── preprocessing/               # Preprocessing components
│   ├── Dockerfile.preprocessing  # Dockerfile for preprocessing
│   ├── preprocessing.py          # Script for preprocessing input data
├── inference/                   # Inference components
│   ├── Dockerfile.infer          # Dockerfile for inference
│   ├── attention_inspect.py      # Script for inspecting attention maps
│   ├── inspect_attention.py      # Main script for running inference and inspection
│   ├── requirements.txt          # Python dependencies for inference
│   ├── utils.py                  # Utility functions for inference
│   ├── vision_transformer.py     # DINO Vision Transformer model implementation
├── docker-compose.yml            # Docker Compose file to run the entire pipeline
├── entrypoint.sh                 # Entrypoint script for Docker container
├── infer.py                      # Main script to run inference outside Docker
├── run_docker.sh                 # Script to run the pipeline using Docker
├── watchdog.py                   # Script to monitor the pipeline
└── .gitignore                    # Git ignore file

Requirements

• Docker
• Docker Compose

Alternatively, you can run the pipeline outside of Docker by installing the required Python packages from respective modules requirements.txt.

Setup

Running with Docker

1. Build and Start the Containers:

   First, ensure you have Docker and Docker Compose installed. Then run the following command to start the pipeline:

   docker-compose up --build

   This will build the Docker containers for preprocessing and inference and start the pipeline.

2. Running the Preprocessing:

   Once the containers are running, you can trigger the preprocessing step by executing:

   docker exec -it preprocessing-container python preprocessing/preprocessing.py

3. Running the Inference:

   After preprocessing, you can run inference by executing:

   docker exec -it inference-container python inference/inspect_attention.py

   This script performs inference using the DINO Vision Transformer model and generates visualizations of attention maps.

Running Without Docker

If you'd prefer to run the pipeline without Docker, you can follow these steps:

1. Install Dependencies:

   Install the required dependencies for inference:

   pip install -r inference/requirements.txt

2. Run Preprocessing:

   Execute the preprocessing script:

   python preprocessing/preprocessing.py

3. Run Inference:

   Execute the inference script:

   python inference/inspect_attention.py

ENV variables

1.	watchdog.py:
•	LOG_DIR (default: "/data/log")
•	INPUT_DIR (default: "/data/sonar")
•	OUTPUT_DIR (default: "/data/processed")
2.	inspect_attention.py:
•	INPUT_DIR (default: "/data/test_imgs")
•	OUTPUT_DIR (default: "/data/inference")
•	LOG_DIR (default: ".")
•	PATCH_SZ (default: 8)
•	ARCH (default: 'vit_small')
•	DOWNSAMPLE_SIZE (default: 5000)
3.	preprocessing.py:
•	INPUT_DIR (default: "/data/processed")
•	OUTPUT_DIR (default: "/data/test_imgs")
•	LOG_DIR (default: ".")
4.	raw.py:
•	INPUT_DIR (default: "/data/sonar")
•	OUTPUT_DIR (default: "/data/processed")
•	LOG_DIR (default: "log")
5.	segmentation.py:
•	INPUT_DIR (default: "/data/sonar")
•	OUTPUT_DIR (default: "/data/processed")
•	LOG_DIR (default: "/data/logs")

Output

The output of the inference step, including generated attention maps and transformed images, will be saved in the inference/output/ directory. Each run will create a timestamped subdirectory for organized output management.

Contributing

Feel free to open issues or submit pull requests if you encounter bugs or have suggestions for improvements.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Acknowledgements

This pipeline uses the DINO Vision Transformer for attention-based image analysis. The implementation is based on research from the original DINO paper by Facebook AI Research (FAIR).