Skip to content

DATASET_SETUP.md

Latest commit

 

History

History
87 lines (63 loc) · 2.81 KB

DATASET_SETUP.md

File metadata and controls

87 lines (63 loc) · 2.81 KB

📁 Flickr8k Dataset Setup Guide

Quick guide to download and set up the Flickr8k dataset for custom training with Snaption.

Note: The dataset is only needed if you want to train your own models. If you're using our pre-trained models, you can skip this entirely!

Quick Setup (5 minutes)

  1. Get Kaggle Account: Sign up at kaggle.com (free)

  2. Download Dataset: Go to Flickr8k on Kaggle

    • Click the "Download" button
    • This downloads archive.zip (~1GB)

  3. Extract and Organize:

    • Create an empty directory named Flickr8k within training/data. The path should be training/data/Flickr8k.
    • Extract the downloaded archive.zip file's contents into this newly created Flickr8k directory.
    • After extraction, the structure should look like this: 
      training/data/Flickr8k/
├── captions.txt
└── Images/
    ├── image1.jpg
    └── ... (8,090 more images)

  4. Verify: Ensure captions.txt and the Images/ folder with .jpg files are present in training/data/Flickr8k/.

  5. Note: Please delete the first line of captions.txt if it contains any header information.

What's Next?

Option A: Use Pre-Trained Models (Recommended)

# Download our trained models instead...
# See: https://github.com/Akhan521/Snaption/releases/tag/v0.1.0

# Quick start with pre-trained model:
pip install -e . # Install snaption package

Option B: Train Your Own Model

# Start training with the dataset you just downloaded:
cd training/
python train.py

🔧 Troubleshooting

"Download failed" or "File not found"

  • Make sure you're logged into Kaggle
  • Try downloading manually from the website
  • Check your internet connection

"No such file or directory"

  • Make sure you extracted the zip file
  • Check that the path training/data/Flickr8k/ exists
  • Verify the Images/ folder contains .jpg files

"Not enough space"

  • The dataset needs ~1GB free space.

📖 Dataset Info

  • Total Size: ~1GB
  • Images: 8,091 photos from Flickr8k
  • Captions: 5 captions per image (total 40,455 captions)
  • License: Research/academic use
  • Quality: Professional photography, diverse scenes

🤔 FAQ

Q: Do I need the dataset to use Snaption?
A: No! Only if you want to train your own models. Use our pre-trained models instead.

Q: Can I use a different dataset?
A: Yes! Just follow the same captions.txt format. See training/dataset.py for details.

Q: Is this dataset free?
A: Yes, it's available for research and educational use.

Need help? Open an issue and I'll assist you!