My Model

🚀 GIT (GenerativeImage2Text), base-sized, fine-tuned on COCO

GIT (GenerativeImage2Text) is a base-sized model fine-tuned on the COCO dataset. It addresses the challenge of generating text from images, offering high - quality image captioning and enabling various vision - language tasks.

🚀 Quick Start

You can use the raw model for image captioning. Check out the model hub to find fine - tuned versions for tasks that interest you. For code examples, refer to the documentation.

✨ Features

• Versatile Task Support: Can be used for image and video captioning, visual question answering (VQA) on images and videos, and even image classification.
• Unique Architecture: A Transformer decoder conditioned on both CLIP image tokens and text tokens, with specific attention mask mechanisms for image and text tokens.

📚 Documentation

Model description

GIT is a Transformer decoder conditioned on both CLIP image tokens and text tokens. The model is trained using "teacher forcing" on a large number of (image, text) pairs.

The goal of the model is to predict the next text token given the image tokens and previous text tokens. When predicting the next text token, the model has full access to the image patch tokens (using a bidirectional attention mask), but only access to the previous text tokens (using a causal attention mask for the text tokens).

This architecture enables the model to handle tasks such as:

• Image and video captioning
• Visual question answering (VQA) on images and videos
• Image classification (by conditioning the model on the image and asking it to generate a class in text)

Intended uses & limitations

You can use the raw model for image captioning. Look for fine - tuned versions on the model hub for tasks that interest you.

Training data

From the paper:

We collect 0.8B image - text pairs for pre - training, which include COCO (Lin et al., 2014), Conceptual Captions (CC3M) (Sharma et al., 2018), SBU (Ordonez et al., 2011), Visual Genome (VG) (Krishna et al., 2016), Conceptual Captions (CC12M) (Changpinyo et al., 2021), ALT200M (Hu et al., 2021a), and an extra 0.6B data following a similar collection procedure in Hu et al. (2021a).

However, this is for the model referred to as "GIT" in the paper, which is not open - sourced. This checkpoint is "GIT - base", a smaller variant of GIT trained on 10 million image - text pairs and then fine - tuned on COCO. See table 11 in the paper for more details.

Preprocessing

Refer to the original repo for details on preprocessing during training. During validation, the shorter edge of each image is resized, followed by center cropping to a fixed - size resolution. Then, frames are normalized across the RGB channels with the ImageNet mean and standard deviation.

Evaluation results

For evaluation results, refer to the paper.

🔧 Technical Details

GIT was introduced in the paper GIT: A Generative Image - to - text Transformer for Vision and Language by Wang et al. and first released in this repository.

Disclaimer: The team releasing GIT did not write a model card for this model, so this model card has been written by the Hugging Face team.

📄 License

This model is released under the MIT license.