Smart Surveillance Using Single Shot MultiBox Detector Implementation with Pytorch

This repo implements SSD (Single Shot MultiBox Detector). The implementation is heavily influenced by the projects ssd.pytorch and Detectron. The design goal is modularity and extensibility.

It has MobileNetV2 SSD/SSD-Lite implementations.

It also has out-of-box support for retraining on Google Open Images dataset.

Dependencies

1. Python 3.6+ (3.6.13)
2. OpenCV
3. Pytorch 1.0 or Pytorch 0.4+
4. Pandas

Initial setup

---

Setup a virtual environment for the installation. Here, Anaconda is used. Run the following commands to install th required dependencies.

conda create -y -n <env-name> python=3.6.13
pip install -r requirements.txt

Run the application

---

To run the application, execute the following command. In case a camera is not used for input, a video file's path can be passed as an argument in the command.

python main.py <optional-video-filepath>

Setup your credentials such as email id and mobile number to receive alerts on threat detection.

MobileNetV2 SSD/SSD-Lite is slower than MobileNetV1 SSD/Lite on PC. However, MobileNetV2 is faster on mobile devices.

Pretrained Models

---

MobileNetV2 SSD-Lite

MB2 Model

Average Precision Across All Classes:0.6860690100560214

Download data

python open_images_downloader.py --root ~/data/open_images --class_names "Handgun,Shotgun" --num_workers 20

It will download data into the folder ~/data/open_images.

The content of the data directory looks as follows.

class-descriptions-boxable.csv       test                        validation
sub-test-annotations-bbox.csv        test-annotations-bbox.csv   validation-annotations-bbox.csv
sub-train-annotations-bbox.csv       train
sub-validation-annotations-bbox.csv  train-annotations-bbox.csv

The folders train, test, validation contain the images. The files like sub-train-annotations-bbox.csv is the annotation file.