salman508/Action_Based_Video_Classification

1import gradio as gr2import os3import numpy as np4import tensorflow as tf5from tensorflow import keras6import pandas as pd7import cv28 9# Define constants10IMG_SIZE = 22411MAX_SEQ_LENGTH = 3012NUM_FEATURES = 204813 14# Load the trained model15model_filepath = "lstm_model.h5"  # Replace with the actual path16loaded_model = keras.models.load_model(model_filepath)17train_df = pd.DataFrame({18    'tag': ['BabyCrawling', 'CricketShot']19})20label_processor = keras.layers.StringLookup(num_oov_indices=0, vocabulary=np.unique(train_df["tag"]))21def crop_center_square(frame):22    y, x = frame.shape[0:2]23    min_dim = min(y, x)24    start_x = (x // 2) - (min_dim // 2)25    start_y = (y // 2) - (min_dim // 2)26    return frame[start_y : start_y + min_dim, start_x : start_x + min_dim]27    28def load_video(path, max_frames=0, resize=(IMG_SIZE, IMG_SIZE)):29    cap = cv2.VideoCapture(path)30    frames = []31    try:32        while True:33            ret, frame = cap.read()34            if not ret:35                break36            frame = crop_center_square(frame)37            frame = cv2.resize(frame, resize)38            frame = frame[:, :, [2, 1, 0]]39            frames.append(frame)40 41            if len(frames) == max_frames:42                break43    finally:44        cap.release()45    return np.array(frames)46# Load the feature extractor47def build_feature_extractor():48    feature_extractor = keras.applications.InceptionV3(49        weights="imagenet",50        include_top=False,51        pooling="avg",52        input_shape=(IMG_SIZE, IMG_SIZE, 3),53    )54    preprocess_input = keras.applications.inception_v3.preprocess_input55 56    inputs = keras.Input((IMG_SIZE, IMG_SIZE, 3))57    preprocessed = preprocess_input(inputs)58 59    outputs = feature_extractor(preprocessed)60    return keras.Model(inputs, outputs, name="feature_extractor")61 62feature_extractor = build_feature_extractor()63 64# Function for preparing a single video for prediction65def prepare_single_video(frames):66    frames = frames[None, ...]67    frame_mask = np.zeros(shape=(1, MAX_SEQ_LENGTH,), dtype="bool")68    frame_features = np.zeros(shape=(1, MAX_SEQ_LENGTH, NUM_FEATURES), dtype="float32")69 70    for i, batch in enumerate(frames):71        video_length = batch.shape[0]72        length = min(MAX_SEQ_LENGTH, video_length)73        for j in range(length):74            frame_features[i, j, :] = feature_extractor.predict(batch[None, j, :])75        frame_mask[i, :length] = 1  # 1 = not masked, 0 = masked76 77    return frame_features, frame_mask78 79# Function for making predictions80def sequence_prediction(video_file):81    class_vocab = label_processor.get_vocabulary()82 83    # Load the video frames84    frames = load_video(video_file)85 86    # Prepare the frames for prediction87    frame_features, frame_mask = prepare_single_video(frames)88 89    # Make predictions using the loaded model90    probabilities = loaded_model.predict([frame_features, frame_mask])[0]91 92    # Get the predicted label93    predicted_label = class_vocab[np.argmax(probabilities)]94 95    return predicted_label96example_list=[97        ["video-1.mp4"],98        ["video-2.mp4"],99    ]100# Gradio interface101iface = gr.Interface(102    fn=sequence_prediction,103    inputs=gr.Video(label="Upload a video file"),104    outputs="text",105    examples=example_list,106)107 108# Launch the Gradio app109iface.launch()