AI-FDU

Project3: Drone Perspective Crowd Counting using Dual Optical Sensors

This repository contains the baseline code for the challenge titled “Drone Perspective Crowd Counting” from the GAIIChallenge hosted at HeyWhale. The objective is to develop an algorithm that accurately counts the number of people from a drone’s perspective using dual optical sensors (presumably RGB and thermal or depth cameras), addressing challenges such as varying densities, occlusions, and altitude-induced perspective changes.

Repository Structure

• model/: Directory to store trained models. Created during setup to separate trained weights and configurations from source code.
• train.py: Script responsible for training the crowd counting model using provided or custom datasets. It implements necessary data preprocessing, model architecture, and training loop.
• test.py: Script designed to evaluate the trained model on a test dataset and generate output predictions, which are then redirected to ans.txt for submission or evaluation purposes.

Quick Start Guide

Setup

1. Clone the Repository:

   git clone https://github.com/AI-FDU/ai-fdu.github.io.git
   

2. Enter Project Directory:

   cd ai-fdu.github.io/pj3
   

3. Create Model Directory (if not done during cloning):

   mkdir model
   

Training

To start the training process, execute:

python train.py

Ensure you have configured the dataset paths and any other necessary parameters within train.py according to your setup.

Testing

Once training is complete, you can test the model on a designated test set by running:

python test.py > ans.txt

This command runs the testing script and redirects its output to ans.txt, which typically contains the predicted counts per image or video frame.

Experimental Requirements

• Develop a solution surpassing or equaling the baseline - 4 points
• Competition results: Determined by ranking and score - 10 points
• Effort and innovation in method: Reflected in the report - 6 points
• Extra points: Pre-presentation plan scheduled for 07/06 - 2 points (as it’s hard to max out)

Final Score: max(20, s1+s2+s3+s4), total of 20 points.

Competition Deadline: June 2, 2024 ‼️ Submit A-board’s best ranking and score screenshots, txt files to eLearning.

Report Deadline: June 14, 2024 ‼️ Submit B-board result screenshots, final code, project report (4 pages), PPT (if applicable) to eLearning.

Submission

Please ensure submission in accordance with the following:

• Develop a solution surpassing or equaling the baseline.
• Provide screenshots of A-board’s best ranking and score, along with txt files.
• Submit B-board result screenshots, final code, project report (4 pages), and PPT (if applicable).
• Submit all materials to eLearning.