formula-one-image-classification-model

Repository holds code and data for image classification model - of cars representing particular teams in Formula 1.

Repository structure

• /assets: place for code to be imported as modules later-on,
• /input: data for model training, validation and testing are stored here. The amount of images commited is an example - I reduced their number to lessen the amount of storage held in GitHub/GitLab,
• /logs: place for logs created during model training,
• /models: space for binary model file (h5 format),
• /requirements: place for files with package requirements to be installed with specific system libraries,
• /static: space for files to be served on a webpage with uvicorn framework,
• /templates: place for website templates to be later-on rendered and served inside an app,
• formula-one-image-classification.ipynb: improved / fixed jupyter notebook explaining model training process (once again, by faw),
• main.py: Python file with API implementation and endpoints created with FastAPI framework,
• README.md: Markdown-based file you are currently reading,
• requirements.txt: file with minimum package requirements necessary for API to work properly,
• environment.yaml: file with full package requirements, necessary for above-mentioned jupyter notebook to work without errors.

Docker image

Building

docker build --no-cache \
--build-arg MINIO_URL="<change_me>" \
--build-arg MINIO_ACCESS_KEY="<change_me>" \
--build-arg MINIO_SECRET_KEY="<change_me>" \
--build-arg DEBUGGING_LOCAL="<change_me>" \
-t f1-image-classification-model:v0.9 -f Dockerfile .

Running

docker run -it \
-e MINIO_URL="<change_me>" \
-e MINIO_ACCESS_KEY="<change_me>" \
-e MINIO_SECRET_KEY="<change_me>" \
-e DEBUGGING_LOCAL="<change_me>" \
f1-image-classification-model:v0.9

Adding image to use in minikube

minikube image load f1-image-classification-model:v0.9

Example

Homepage

Predict page

Image

Response output

{
    "prediction": {
        "class": "mclaren",
        "confidence_percent": 99.96,
        "message": "1st Prediction: mclaren with 99.96% confidence.",
    },
    "predictions": {
        1: {
            "class": "mclaren",
            "confidence_percent": 99.96,
            "message": "1st Prediction: mclaren with 99.96% confidence.",
        },
        2: {
            "class": "bwt",
            "confidence_percent": 0.04,
            "message": "2nd Prediction: bwt with 0.04% confidence.",
        },
        3: {
            "class": "toro_rosso",
            "confidence_percent": 0.0,
            "message": "3rd Prediction: toro_rosso with 0.00% confidence.",
        },
        4: {
            "class": "williams",
            "confidence_percent": 0.0,
            "message": "4th Prediction: williams with 0.00% confidence.",
        },
        5: {
            "class": "mercedes",
            "confidence_percent": 0.0,
            "message": "5th Prediction: mercedes with 0.00% confidence.",
        },
        6: {
            "class": "haas",
            "confidence_percent": 0.0,
            "message": "6th Prediction: haas with 0.00% confidence.",
        },
        7: {
            "class": "redbull",
            "confidence_percent": 0.0,
            "message": "7th Prediction: redbull with 0.00% confidence.",
        },
        8: {
            "class": "alfa_romeo",
            "confidence_percent": 0.0,
            "message": "8th Prediction: alfa_romeo with 0.00% confidence.",
        },
        9: {
            "class": "ferrari",
            "confidence_percent": 0.0,
            "message": "9th Prediction: ferrari with 0.00% confidence.",
        },
        10: {
            "class": "renault",
            "confidence_percent": 0.0,
            "message": "10th Prediction: renault with 0.00% confidence.",
        },
    },
}

Model structure

Model: "sequential"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ conv2d (Conv2D)                 │ (None, 254, 254, 16)   │           448 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d (MaxPooling2D)    │ (None, 127, 127, 16)   │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_1 (Conv2D)               │ (None, 125, 125, 32)   │         4,640 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_1 (MaxPooling2D)  │ (None, 62, 62, 32)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_2 (Conv2D)               │ (None, 60, 60, 16)     │         4,624 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ max_pooling2d_2 (MaxPooling2D)  │ (None, 30, 30, 16)     │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ flatten (Flatten)               │ (None, 14400)          │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense (Dense)                   │ (None, 256)            │     3,686,656 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_1 (Dense)                 │ (None, 10)             │         2,570 │
└─────────────────────────────────┴────────────────────────┴───────────────┘
 Total params: 3,698,938 (14.11 MB)
 Trainable params: 3,698,938 (14.11 MB)
 Non-trainable params: 0 (0.00 B)

Sources / Acknowledgements

F1 Cars

classification model: by faw, available here,

images dataset: by Sérgio Gomes and José Henrique Brito, available here,

containerization and API creation: by Aleksander Zawalich, available here.