vehicles-visual-signals-processing

Project for vehicles signals recognition for Master Degree Paper, Russia, Moscow, MISIS 2021-2022

- Python version: 3.9.10

• Annotation tool: CVAT
• Project is CLOSED

USAGE

There are 2 types of NN in this repository:

1. CNN for single image classification - train_cnn_model.py
2. LSTM for sequence classification - train_lstm_model.py

HINT: use key -h to check arguments to check parameters of learning

DATASETS

• Archive with 3 datasets

1. Single images (854 items)
2. Difference betwen first and last frame (838 items)
3. Sequence of 5 image as AVI files (807 items)

MODEL RESULTS

1. LSTM results
2. CNN with single images
3. CNN with difference

DATASET COOKBOOK:

To start video chunks from video:

usage: dataset_generator.py [-h] [-i INPUT] [-o OUTPUT] [-m {sequence,singleshot}] [--overwrite] [--debug]

optional arguments: -h, --help show this help message and exit

-i INPUT, --input INPUT Input directory with videos and annotation archive

-o OUTPUT, --output OUTPUT Output directory for dataset

-m {sequence,singleshot}, --mode {sequence,singleshot} Dataset generator mode. Sequence for MJPG and singleshot for JPG

--overwrite Overwrite current dataset directory if exists

--debug Enable debug log writing

Example: raw_data

Extractor description:

1. Default directory for input video files: "data\raw_data" (can be changed in .\utils\constants.py)
2. Natively supported video file format:

• ".TS" with 30 FPS.
• Example: "REC25915.ts"

3. Annotation support type:

• XML (CVAT v.1.1)
• Must be track with 2 points
• Must be placed "as is" in native ZIP archive from CVAT export
• Must be named natively with CVAT export name
• Example: "task_rec25915.ts_cvat for video 1.1.zip"

4. Extracted video chunks are placed in new dataset directory
5. Extractor creates MJPG chunk from each availible track annotation:

• format: "{file}{label_name}{class_type}_{class_name}_tr{track_num}_seq{chunk_num}_fr{frame_num}.mjpg"

HINTS:

1. Creating virtual environment:

• python -m venv .venv

2. Activating virtual environment:

• CMD: ..venv\Scripts\activate.bat
• PowerShell: ..venv\Scripts\Activate.ps1
• Linux Terminal: source .venv/bin/activate

3. Installation requirements:

• pip3 install -r requirements.txt

4. Installation CUDA for Windows

5. Installation PyTorch docs