BionicEye Agent

This repository contains code for a Reinforcement Learning agent using a Vision Transformer to solve image classification tasks by active visual exploration.

---

Quick Start

1. Local Machine

All commands are run from the project root directory.

# 1. Create virtual environment and activate it
python3 -m venv venv
source venv/bin/activate

# 2. Install dependencies
pip install -r requirements.txt

# 3. Start training
python current_main_script.py

2. Remote Server (SC-09)

Step A: Copy files to server (run on your local machine)

This command copies your current project folder to the server. You will be prompted for the password.

rsync -avz \
  --exclude 'venv' \
  --exclude 'checkpoints' \
  --exclude 'visualizations' \
  --exclude 'wandb' \
  --exclude '.git' \
  --exclude '.gitignore' \
  --exclude '__pycache__' \
  --exclude '*.npz' \
  -e "ssh" \
  . user9@87.242.102.117:~/BionicEye

Step B: Setup and Run (run on the remote server)

Connect to the server (ssh user9@87.242.102.117), then run these commands.

# --- First-time setup on server (only need to do this once) ---
cd ~/BionicEye
sudo apt install python3.10-venv -y
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
# --- End of setup ---

Step C: Check Server Load (Important!)

Before starting a new training, check if the GPUs are already in use by someone else.

nvidia-smi

Look at the process list at the bottom. If you see other python processes consuming significant GPU memory, it's best to wait or coordinate with your colleagues.

Step D: Start Training

If the server is free and all dependencies are installed, start training interactively.

# Connect to server and start training directly
ssh -t user9@87.242.102.117 "cd ~/BionicEye && source venv/bin/activate && python current_main_script.py"

Alternative: Use screen for long-running sessions

If you want to use screen (for sessions you can detach/reattach), start it interactively:

# Connect to server
ssh user9@87.242.102.117

# Create screen session
screen -S training

# Inside screen: activate environment and start training
cd ~/BionicEye
source venv/bin/activate
python current_main_script.py

# To detach from screen (training continues): Ctrl+A, then D
# To reattach later: screen -r training

Screen commands:

• Ctrl+A, D — detach from session (training continues in background)
• screen -r training — reattach to the training session
• screen -ls — list all active screen sessions
• screen -S training -X quit — kill the training session

You can safely disconnect from SSH after detaching from screen. The training will continue running on the server.

Step E: Downloading Results from Server

To copy the generated checkpoints or visualizations from the server back to your local machine, run these commands on your local terminal.

# To download the latest checkpoint:
scp user9@87.242.102.117:~/BionicEye/checkpoints/latest_checkpoint.pth ./checkpoints/

# To download all visualizations:
scp -r user9@87.242.102.117:~/BionicEye/visualizations/ ./visualizations/