KAGGLE_FIX.md

Kaggle Read-Only File System Fix

Problem

OSError: [Errno 30] Read-only file system: '/kaggle/input/yolo-mpeb-training-code/code/datasets'

Root Cause

In Kaggle:

• /kaggle/input/ is READ-ONLY (contains your uploaded datasets)
• /kaggle/working/ is WRITABLE (for outputs and temporary files)

The dataset YAML was trying to download/create files in /kaggle/input/, which is not allowed.

Solution

✅ Fixed Files

1. dataset_example.yaml - Changed dataset path

   # Before (WRONG):
   path: VisDrone
   
   # After (CORRECT):
   path: /kaggle/working/VisDrone
   

2. train_kaggle.py - New Kaggle-specific training script

   • Properly handles Kaggle paths
   • Copies files from /kaggle/input/ to /kaggle/working/
   • Sets up training in writable directory

3. kaggle_training_notebook.ipynb - Ready-to-use Kaggle notebook

   • Complete training workflow
   • Validation and testing cells
   • Visualization of results

4. KAGGLE_SETUP.md - Comprehensive setup guide

   • Step-by-step instructions
   • Troubleshooting tips
   • Path explanations

How to Use

Option 1: Use the Notebook (Recommended)

1. Upload all files to a Kaggle dataset
2. Create a new Kaggle notebook
3. Add your dataset as input
4. Upload kaggle_training_notebook.ipynb
5. Run all cells

Option 2: Use the Python Script

1. Upload all files to a Kaggle dataset
2. Create a new Kaggle notebook
3. Run:

   import shutil
   shutil.copy('/kaggle/input/yolo-mpeb-training-code/code/train_kaggle.py', 
               '/kaggle/working/train_kaggle.py')
   !python /kaggle/working/train_kaggle.py
   

Key Changes Summary

File	Change	Reason
dataset_example.yaml	path: VisDrone → path: /kaggle/working/VisDrone	Use writable directory
train_kaggle.py	New file	Kaggle-specific paths and setup
kaggle_training_notebook.ipynb	New file	Easy-to-use notebook template
KAGGLE_SETUP.md	New file	Documentation and troubleshooting

Verification

After the fix, training should start successfully:

Ultralytics 8.3.239 🚀 Python-3.11.13 torch-2.6.0+cu124 CUDA:0 (Tesla P100-PCIE-16GB, 16269MiB)
engine/trainer: ...
Downloading VisDrone dataset to /kaggle/working/VisDrone...
Training starting...

Important Notes

1. Dataset Download: First run will download ~2.3 GB VisDrone dataset
2. Training Time: ~6-8 hours on Tesla P100
3. Save Outputs: Download weights before closing notebook
4. GPU Required: Enable GPU in Kaggle settings

Files to Upload to Kaggle Dataset

Upload these files to your Kaggle dataset:

• ✅ yolov8_mpeb.yaml - Model architecture
• ✅ yolov8_mpeb_modules.py - Custom modules
• ✅ dataset_example.yaml - Dataset config (FIXED)
• ✅ train_kaggle.py - Training script (NEW)

Quick Test

To verify the fix works, run this in a Kaggle notebook:

import yaml
with open('/kaggle/input/yolo-mpeb-training-code/code/dataset_example.yaml') as f:
    config = yaml.safe_load(f)
    print(f"Dataset path: {config['path']}")
    # Should print: /kaggle/working/VisDrone

Support

If you still get errors:

1. Check that dataset path is /kaggle/working/VisDrone
2. Verify GPU is enabled
3. Ensure all files are in your Kaggle dataset
4. Check the KAGGLE_SETUP.md for detailed troubleshooting