Faster R-CNN Training Repository

This repository contains implementation of Faster R-CNN using PyTorch 2.3.0

Setup

Prerequisites

• Docker
• GPU with CUDA support
• Change all kinds of hyperparameters using the config.yaml ONLY.
• Follow autotrain.sh file to run the repo

Usage

Docker creation

Use image and container name as per your need.

To create a new image use this command :-

• sudo docker build -t image .

NOTE

This code base can auto detect GPU given to it. So only enter those GPUs that you want to use, in the below command

Create Docker container for this using :-

• sudo docker run -it --name container_name -v /path/to/code:/localizer --gpus '"device=0,1,2,3"' -u $(id -u):$(id -g) --shm-size 16G image

Workflow Design (TRAINING & INFERENCING)

• UPDATE FILE PATHS, COLUMN NAMES IN CONFIG.YAML
• REFER TO autotrain.sh for steps to start training.
• ADJUST PATHS AS NEEDED.

Repository Structure

• configs/: Contains configuration files config.yaml

• models/checkpoints/: Directory for saving and loading model checkpoints.

• script/: Contains the main scripts:

• train_frcnn.py: For training the Faster R-CNN model.

• accuracy.py: For evaluating model accuracy

• data_split.py: For splitting datasets.

• dataloader.py: For loading datasets.

• plot.py and plotter.py: For visualizing results.

• preprocess_data_csv.py: For preprocessing data in CSV format.

• .gitignore: Specifies files and directories to ignore in version control.

• Dockerfile: Docker configuration for containerized environments.

• requirements.txt: Lists Python dependencies.

• README.md: Documentation for the repository.