Official implementation and dataset for the NAACL 2024 paper "ComCLIP: Training-Free Compositional Image and Text Matching"

Download flickr8k, flickr30k image caption datasets

A deep learning model that generates descriptions of an image.

Image captioning model with Resnet50 encoder and LSTM decoder

PyTorch implementation of 'CLIP' (Radford et al., 2021) from scratch and training it on Flickr8k + Flickr30k

Visual Elocution Synthesis

Karpathy Splits json files for image captioning

ImgCap is an image captioning model designed to automatically generate descriptive captions for images. It has two versions CNN + LSTM model and CNN + LSTM + Attention mechanism model.

Preprocess the Flickr30k dataset

Processing data produced by flickr30k_entities to use as regional description for densecap model

Image captioning generation using Swin transformer and GRU attention mechanism

Image captioning model using EfficientNetB0 as encoder and a custom Transformer decoder, trained on the Flickr30k dataset. Demonstrates full model architecture, preprocessing, and BLEU-based evaluation in TensorFlow. Built as an educational resource to explain Transformer architecture step-by-step.

Implementation of CLIP from OpenAI using pretrained Image and Text Encoders.

Attention Based image captioning

"Flickr30k_image_captioning" is a project or repository focused on image captioning using the Flickr30k dataset. The project aims to develop and showcase algorithms and models that generate descriptive captions for images.

This project aims to provide real-time visual-to-audio conversion, empowering visually impaired users by describing images through generated captions and synthesized audio. The system employs a Transformer-based image captioning model and integrates with a browser extension for seamless functionality.