Add files via upload

yolo_cnn_lstm/test_cnn_lstm.ipynb

@@ -0,0 +1,91 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Using CNN-LSTM to classify violent clips"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "import cv2\n",
+    "import numpy as np\n",
+    "import tensorflow as tf\n",
+    "\n",
+    "# The video should be converted to frames first. During training, three frames are extracted per second,\n",
+    "# each frame is resized to 64x64 pixels, and the LSTM layer takes 10 consecutive frames as input.\n",
+    "def preprocess_video(video_path, frame_count=10, frame_size=(64, 64)):\n",
+    "    cap = cv2.VideoCapture(video_path)\n",
+    "    frames = []\n",
+    "    while len(frames) < frame_count:\n",
+    "        ret, frame = cap.read()\n",
+    "        if not ret:\n",
+    "            break\n",
+    "        frame = cv2.resize(frame, frame_size)\n",
+    "        frames.append(frame)\n",
+    "    cap.release()\n",
+    "    \n",
+    "    if len(frames) == 0:\n",
+    "        return np.zeros((frame_count, frame_size[0], frame_size[1], 3))\n",
+    "    elif len(frames) < frame_count:\n",
+    "        frames.extend([np.zeros_like(frames[0])]*(frame_count - len(frames)))\n",
+    "    return np.array(frames)\n",
+    "\n",
+    "def load_test_data(folder_path, frame_count=10, frame_size=(64, 64)):\n",
+    "    test_data = []\n",
+    "    video_files = [f for f in os.listdir(folder_path) if f.endswith('.mp4') or f.endswith('.avi')]\n",
+    "    for video_file in video_files:\n",
+    "        video_path = os.path.join(folder_path, video_file)\n",
+    "        frames = preprocess_video(video_path, frame_count, frame_size)\n",
+    "        test_data.append(frames)\n",
+    "    return np.array(test_data), video_files\n",
+    "\n",
+    "# Folder containing your test videos\n",
+    "test_video_folder = \"D:/Yolo/test/result\"\n",
+    "test_data, test_video_files = load_test_data(test_video_folder)\n",
+    "test_data = test_data / 255.0\n",
+    "\n",
+    "# This model can achieve over 90% accuracy on the training set and 77% on the validation set.\n",
+    "model_path = \"D:/CNN-LSTM/models/violence_detection_model_conv_64_lstm_64.h5\"\n",
+    "model = tf.keras.models.load_model(model_path, compile=False)\n",
+    "\n",
+    "# Recompile the model with a compatible optimizer and loss function\n",
+    "model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])\n",
+    "\n",
+    "predictions = model.predict(test_data)\n",
+    "predicted_classes = np.argmax(predictions, axis=1)\n",
+    "\n",
+    "class_names = [\"high-level violence\", \"low-level violence\", \"non-violence\"]\n",
+    "for video_file, predicted_class in zip(test_video_files, predicted_classes):\n",
+    "    print(f\"Video: {video_file}, Predicted Class: {class_names[predicted_class]}\")"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "test",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.8.18"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

yolo_cnn_lstm/yolo.ipynb

@@ -0,0 +1,160 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# 1. Training the Yolo model"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import ultralytics\n",
+    "ultralytics.checks()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from ultralytics import YOLO\n",
+    "\n",
+    "from IPython.display import display, Image"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!yolo task=detect mode=train model=yolov8s.pt data=Violence.yaml epochs=50 imgsz=416 plots=True"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# 2. Using the YOLO model to filter violent clips"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import cv2\n",
+    "import os\n",
+    "from ultralytics import YOLO\n",
+    "\n",
+    "# Load the trained YOLOv8 model\n",
+    "model = YOLO('D:\\\\Yolo\\\\runs\\\\detect\\\\train\\\\weights\\\\best.pt')\n",
+    "\n",
+    "# Input video file path\n",
+    "input_video_path = 'D:\\\\Yolo\\\\test\\\\videos'\n",
+    "output_clip_path = 'D:\\\\Yolo\\\\test\\\\result'\n",
+    "\n",
+    "# Create the output folder (if it doesn't exist)\n",
+    "os.makedirs(output_clip_path, exist_ok=True)\n",
+    "\n",
+    "# Get all video files\n",
+    "video_files = [f for f in os.listdir(input_video_path) if f.endswith('.mp4')]\n",
+    "\n",
+    "for video_file in video_files:\n",
+    "    cap = cv2.VideoCapture(os.path.join(input_video_path, video_file))\n",
+    "    fps = cap.get(cv2.CAP_PROP_FPS)\n",
+    "    frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))\n",
+    "    frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))\n",
+    "\n",
+    "    frame_count = 0\n",
+    "    clip_number = 1\n",
+    "    clip_frames = []\n",
+    "    start_time = None\n",
+    "    last_detection_time = None\n",
+    "\n",
+    "    while cap.isOpened():\n",
+    "        ret, frame = cap.read()\n",
+    "        if not ret:\n",
+    "            break\n",
+    "\n",
+    "        # Perform object detection on the current frame\n",
+    "        results = model(frame)\n",
+    "\n",
+    "        # Extract detection results\n",
+    "        detections = results[0].boxes.data.cpu().numpy()\n",
+    "        current_time = frame_count / fps\n",
+    "        if len(detections) > 0:\n",
+    "            # If an object is detected, save the current frame\n",
+    "            clip_frames.append(frame)\n",
+    "            if start_time is None:\n",
+    "                start_time = current_time\n",
+    "            last_detection_time = current_time\n",
+    "        else:\n",
+    "            # If no object is detected but the time interval is less than 2 seconds, save the current frame\n",
+    "            if last_detection_time is not None and current_time - last_detection_time < 2:\n",
+    "                clip_frames.append(frame)\n",
+    "            else:\n",
+    "                # If the time interval is greater than 2 seconds and there are accumulated frames, save these frames as a video clip\n",
+    "                if len(clip_frames) > 0:\n",
+    "                    end_time = current_time if last_detection_time is None else last_detection_time\n",
+    "                    start_time_formatted = f'{int(start_time // 60):02d}_{int(start_time % 60):02d}'\n",
+    "                    end_time_formatted = f'{int(end_time // 60):02d}_{int(end_time % 60):02d}'\n",
+    "                    clip_output_path = os.path.join(output_clip_path, f'{os.path.splitext(video_file)[0]}_clip_{clip_number}_{start_time_formatted}_to_{end_time_formatted}.mp4')\n",
+    "                    out = cv2.VideoWriter(clip_output_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (frame_width, frame_height))\n",
+    "                    for clip_frame in clip_frames:\n",
+    "                        out.write(clip_frame)\n",
+    "                    out.release()\n",
+    "                    clip_frames = []\n",
+    "                    clip_number += 1\n",
+    "                    start_time = None\n",
+    "                    last_detection_time = None\n",
+    "\n",
+    "        frame_count += 1\n",
+    "\n",
+    "    # Process the remaining frames at the end of the video\n",
+    "    if len(clip_frames) > 0:\n",
+    "        end_time = current_time if last_detection_time is None else last_detection_time\n",
+    "        start_time_formatted = f'{int(start_time // 60):02d}_{int(start_time % 60):02d}'\n",
+    "        end_time_formatted = f'{int(end_time // 60):02d}_{int(end_time % 60):02d}'\n",
+    "        clip_output_path = os.path.join(output_clip_path, f'{os.path.splitext(video_file)[0]}_clip_{clip_number}_{start_time_formatted}_to_{end_time_formatted}.mp4')\n",
+    "        out = cv2.VideoWriter(clip_output_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (frame_width, frame_height))\n",
+    "        for clip_frame in clip_frames:\n",
+    "            out.write(clip_frame)\n",
+    "        out.release()\n",
+    "\n",
+    "    cap.release()\n",
+    "\n",
+    "cv2.destroyAllWindows()\n",
+    "\n"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "VD",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.9.19"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}