Blip Base Captioning Ft Hl Actions

🚀 BLIP-base fine-tuned for Image Captioning on High-Level descriptions of Actions

This project presents a fine-tuned BLIP base model on the HL dataset for action generation of images. It aims to generate high - level descriptions of actions in images.

✨ Features

• Fine - tuned on HL dataset: Trained specifically for action generation in images using the HL dataset.
• Optimized training: Utilizes Adam optimizer, half - precision (fp16), and a specific learning rate for 6 epochs.
• Multiple evaluation metrics: Evaluated using Cider, SacreBLEU, and Rouge - L.

📦 Installation

Since the library used is transformers, you can install it using the following command:

pip install transformers

💻 Usage Examples

Basic Usage

import requests
from PIL import Image
from transformers import BlipProcessor, BlipForConditionalGeneration

processor = BlipProcessor.from_pretrained("michelecafagna26/blip-base-captioning-ft-hl-actions")
model = BlipForConditionalGeneration.from_pretrained("michelecafagna26/blip-base-captioning-ft-hl-actions").to("cuda")

img_url = 'https://datasets-server.huggingface.co/assets/michelecafagna26/hl/--/default/train/0/image/image.jpg' 
raw_image = Image.open(requests.get(img_url, stream=True).raw).convert('RGB')


inputs = processor(raw_image, return_tensors="pt").to("cuda")
pixel_values = inputs.pixel_values

generated_ids = model.generate(pixel_values=pixel_values, max_length=50,
            do_sample=True,
            top_k=120,
            top_p=0.9,
            early_stopping=True,
            num_return_sequences=1)

processor.batch_decode(generated_ids, skip_special_tokens=True)

>>> "she is holding an umbrella."

🔧 Technical Details

Model fine-tuning 🏋️‍

• Epochs: The model was trained for 6 epochs.
• Learning rate: A learning rate of 5e−5 was used.
• Optimizer: The Adam optimizer was employed.
• Precision: Half - precision (fp16) was used during training.

Test set metrics 🧾

📄 License

This project is licensed under the Apache 2.0 license.

📚 Documentation

BibTex and citation info

@inproceedings{cafagna2023hl,
  title={{HL} {D}ataset: {V}isually-grounded {D}escription of {S}cenes, {A}ctions and
{R}ationales},
  author={Cafagna, Michele and van Deemter, Kees and Gatt, Albert},
  booktitle={Proceedings of the 16th International Natural Language Generation Conference (INLG'23)},
address = {Prague, Czech Republic},
  year={2023}
}