# Hindi Image Captioning Model

This is an encoder-decoder image captioning model made with VIT encoder and GPT2-Hindi as a decoder. This is a first attempt at using ViT + GPT2-Hindi for image captioning task. We used the Flickr8k Hindi Dataset available on kaggle to train the model.

This model was trained using HuggingFace course community week, organized by Huggingface.

## How to use

Here is how to use this model to caption an image of the Flickr8k dataset:
```python
import torch
import requests
from PIL import Image
from transformers import ViTFeatureExtractor, AutoTokenizer, VisionEncoderDecoderModel

if torch.cuda.is_available():
    device = 'cuda'
else:
    device = 'cpu'

url = 'https://shorturl.at/fvxEQ'
image = Image.open(requests.get(url, stream=True).raw)

encoder_checkpoint = 'google/vit-base-patch16-224'
decoder_checkpoint = 'surajp/gpt2-hindi'

feature_extractor = ViTFeatureExtractor.from_pretrained(encoder_checkpoint)
tokenizer = AutoTokenizer.from_pretrained(decoder_checkpoint)
model = VisionEncoderDecoderModel.from_pretrained('team-indain-image-caption/hindi-image-captioning').to(device)

#Inference
sample = feature_extractor(image, return_tensors="pt").pixel_values.to(device)
clean_text = lambda x: x.replace('<|endoftext|>','').split('\n')[0]

caption_ids = model.generate(sample, max_length = 50)[0]
caption_text = clean_text(tokenizer.decode(caption_ids))
print(caption_text)
```

## Training data
We used the Flickr8k Hindi Dataset, which is the translated version of the original Flickr8k Dataset, available on Kaggle to train the model.

## Training procedure
This model was trained during HuggingFace course community week, organized by Huggingface. The training was done on Kaggle GPU.