| --- |
| license: apache-2.0 |
| pipeline_tag: image-classification |
| --- |
| # Screenshot classifier |
|
|
| A lightweight binary classifier for screenshot detection. |
|
|
| ## Model |
|
|
| - `mobilenetv3_small_screenshot.keras` — trained Keras model for screenshot / not_screenshot classification |
| |
| ## Usage |
| |
| ```python |
| import tensorflow as tf |
| from tensorflow import keras |
| |
| model = keras.models.load_model("mobilenetv3_small_screenshot.keras") |
|
|
| # Example inference |
| img = keras.utils.load_img("image.png", target_size=(224, 224)) |
| arr = keras.utils.img_to_array(img) |
| arr = tf.keras.applications.mobilenet_v3.preprocess_input(arr) |
| arr = arr[None, ...] |
| pred = model.predict(arr) |
| score = float(pred[0, 0]) |
| label = "screenshot" if score >= 0.5 else "not_screenshot" |
| print(label, score) |
| ``` |
| |
| ## Input format |
| |
| - RGB images |
| - resized to `224x224` |
| - model expects a single sigmoid score |
| |
| ## Prediction threshold |
| |
| - `score >= 0.5` → `screenshot` |
| - `score < 0.5` → `not_screenshot` |
|
|
| ## Notes |
|
|
| - Model backbone: `MobileNetV3Small` pretrained on ImageNet |
| - Output activation: `sigmoid` |
| - Loss used during training: `binary_crossentropy` |
|
|
